opencompany
CHANGELOG.md
Keep a Changelog 1.1.0

Changelog

All notable changes to this project will be documented in this file. The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

Latest 1.31.0Released 2026-09-1851 entries total
1.31.02026-09-18
Added4 entries
  • turn on approve for me in preferences → permissions to let routine plugin actions continue without waiting for you. it can approve searches, document edits, and requested messages that help with your task. destructive, paid, high-risk, or uncertain actions still ask when set to ask; your saved tool permissions stay the same (#1944, #1945) — @louis.
  • try the company agents beta in preferences to give recurring work to a named agent. set its instructions, triggers, and slack identity, then follow its work in a shared run history. teammates can run it and read results; only its owner can change it, and runs use the owner's connected accounts (#1951) — @louis.
  • attach or paste up to five screenshots when sending feedback. review the previews, then send with cmd/ctrl+enter (#1948) — @louis.
  • see when a reply is ready without reopening the app. the browser tab shows done when your open conversation finishes while you are in another tab (#1947) — @louis.
Changed2 entries
  • ask chat to run up to four different workflows in one message, with a separate task for each. name the workflows in your request; the composer still accepts one workflow mention at a time (#1938) — @louis.
  • dark mode now uses charcoal gray backgrounds instead of green-tinted ones (#1942) — @louis.
Fixed6 entries
  • completed coding chats now clear their working indicator, timer, and stop button without a page reload. finished tool activity also collapses so the answer is easier to read (#1946, #1949) — @louis.
  • stopping a chat now shows stopping immediately and clears the working state once the run ends (#1939) — @louis.
  • keep line breaks added with shift+enter when you continue typing in the composer (#1943) — @louis.
  • continue long chats and tasks more reliably, including after sharing screenshots. older images can now inform the conversation summary instead of causing follow-ups to fail (#1927, #1932) — @louis.
  • ask codex to read a saved workflow's instructions and settings again without the request failing (#1928) — @louis.
  • coding tasks now show the model they actually used. the sandbox badge reports whether the machine is running, asleep, or deleted, and shows unknown if its status cannot be checked (#1936, #1941) — @louis.
1.30.02026-09-17
Added1 entry
  • workspace admins can now create and manage workflows from chat. describe the work and when it should run, then ask to edit, pause, run, or archive it. create one-step workflows with a schedule or manual start; use the editor for more steps, event triggers, model choices, and channels (#1926) — @louis.
Changed5 entries
  • new coding chats remember your last model and reasoning effort in this browser, separately for codex and claude code. existing chats keep their own choices (#1920, #1923) — @louis.
  • connecting a codex subscription as a workspace admin now turns on sharing with your team automatically (#1911) — @louis.
  • plugin and repository mentions are easier to read, with service icons and colors, clearer contrast in sent messages, and labels that line up with the surrounding text (#1910, #1915, #1918) — @louis.
  • the composer now says queue a follow-up while a coding agent works, so it is clear your next message will wait for the current turn (#1922) — @louis.
  • the optional connect mcp prompt no longer appears in the sidebar. find the setup guide in settings → mcp when you need it (#1931) — @louis.
Fixed7 entries
  • finished tasks now stop showing a working indicator and running timer, even when part of the page is slow to update (#1933) — @louis.
  • codex can now ask questions with suggested answers during normal coding chats and tasks. pending questions open automatically so you can answer and let work continue (#1930, #1935) — @louis.
  • replying in a slack thread now resumes a paused task instead of leaving your reply waiting indefinitely (#1925) — @louis.
  • personal workflows can use your personal skills again, including scheduled runs that previously failed before returning a result (#1917) — @louis.
  • connecting a plugin during signup now installs it and opens its account connection in one click (#1916) — @louis.
  • github access in active coding sessions no longer breaks when another session refreshes the shared connection (#1913) — @louis.
  • use your mac's two-finger swipe to go back and forward in the app again (#1929) — @louis.
1.29.02026-09-16
Added8 entries
  • message the opencompany bot in slack to start a task with your own connected tools and context. keep replying in its thread to continue. your slack email must match your account in the connected workspace; admins may need to reconnect slack in settings → channels (#1900) — @louis.
  • try the imessage assistant beta to text opencompany from your phone. turn it on in preferences → beta features, then link your number in settings → channels → imessage. it can search the web, use your wiki, skills, and connected tools; actions that need approval cannot run over text (#1903) — @louis.
  • let workflows remember what happened last time. for workflows using chat models, turn on memory under advanced to carry a note between runs. you can read or clear it there; memory starts off (#1887) — @louis.
  • start a workflow when an email arrives. enable email received in the gmail plugin, then add it as a workflow trigger and optionally filter by label. mail is checked every five minutes; up to 25 messages can trigger workflows per check, and extra matches are skipped (#1893) — @louis.
  • start a workflow from a posthog event, such as a signup. connect your project with a read-only personal api key in the plugin's events setup, then choose the event in your workflow trigger (#1906) — @louis.
  • connect todoist to find, create, and update tasks and projects from chat. changes ask for approval by default; permanent deletion starts off (#1878) — @louis.
  • see a workflow's past runs on its detail page, with status, outcome, duration, and a link to the task. open all runs to see its full history (#1884) — @louis.
  • type @ to add a link to an installed plugin or accessible github repository in chat or workflow instructions. these references help name what you mean; they do not grant access or change tool permissions (#1907) — @louis.
Changed5 entries
  • choose whether each workflow can post to slack in its channels section. give its posts a display name and upload an avatar; an admin may need to reconnect slack to enable the custom identity (#1886, #1892, #1901) — @louis.
  • workflows can put a short slack update in the channel and details in its thread. replies now get a progress reaction while work runs, then an answered or needs-attention mark. reconnect older slack installs to enable reactions (#1895, #1897, #1898) — @louis.
  • search the same model choices in workflows as in chat, including qwen 3.8 max, deepseek v4, and grok 4.6. coding agents have their own tab, and the composer shows when a workflow uses a cloud sandbox (#1905, #1904) — @louis.
  • see a coding chat or task's pull request status in the sidebar. running, waiting, and unread indicators take priority until you have seen the result (#1891, #1896, #1899, #1908) — @louis.
  • tool activity and approval cards now name the service and action, so you can see which connected app the agent is using (#1885) — @louis.
Fixed3 entries
  • connected tools no longer reject unrelated actions as duplicates after a connection restarts (#1879) — @louis.
  • sending a follow-up to a task no longer adds an unnecessary five-second wait before confirming it was sent (#1882) — @louis.
  • workflows with slack enabled can now use the workspace bot consistently across chat models and coding agents (#1890) — @louis.
Removed1 entry
  • routines, the old scheduled tasks, no longer run. recreate recurring work as a workflow with a schedule trigger. tasks from past routine runs remain available (#1902) — @louis.
Security1 entry
  • personal workflow schedules are now visible only to their creator. switching a company workflow to personal also removes teammates' access to its schedule (#1894) — @louis.
1.28.02026-09-15
Added4 entries
  • send a follow-up while a coding chat or task is working. it queues for the next turn; choose steer to redirect the current turn, or remove it if you change your mind. if the agent cannot take the instruction mid-turn, it stays queued (#1866, #1873) — @louis.
  • find prospects, research people at a company, and look up work emails with the lead research plugin. prices are shown before you spend, and workspace admins can set a daily limit. install the plugin to keep using prospecting tools previously available through capabilities (#1864) — @louis.
  • start a workflow when a convex function fails, with the function and error as context. enable error events in the convex plugin, then choose function failed as the trigger. repeated matching errors are grouped into 15-minute windows. requires convex pro; convex charges for log streams (#1862) — @louis.
  • choose gpt 5.6 sol or terra for workflow steps. models covered by your workspace's shared subscription now show included in the picker (#1877) — @louis.
Changed6 entries
  • connect a coding subscription and install plugins during workspace setup, so your first chat has the tools you need (#1876) — @louis.
  • open plugins and skills from the main sidebar while keeping your chats, tasks, and wiki in view. both have moved out of settings; old links still work (#1855, #1881) — @louis.
  • choose the size of new cloud coding sessions in settings → sandboxes. workspace admins can compare machine sizes and hourly costs there (#1859) — @louis.
  • chat and tasks can now make up to 32 connected-tool calls per turn, up from 16, so longer research and updates can finish before hitting the limit (#1861) — @louis.
  • keep workflow instructions in the steps. event triggers no longer show an additional run context field; existing saved context is preserved (#1865) — @louis.
  • running subagents start collapsed to keep the conversation readable. expand a row to follow its work (#1836) — @louis.
Fixed9 entries
  • coding sessions now finish or stop correctly after a bug left them showing as running even when the agent was done. resumed sessions also avoid repeating earlier tool activity (#1875, #1867) — @louis.
  • plugin tool permissions now keep showing your saved on, ask, or off choice instead of appearing to revert. connected accounts also update reliably when you have a custom mcp connection (#1872, #1858) — @louis.
  • slack follow-ups now include the thread history and readable sender details. workflows send replies back to the original thread explicitly, so an internal task result is no longer automatically copied to slack (#1851) — @louis.
  • workflow tasks now show their actual instructions or event context as the opening request instead of a generic run this workflow message (#1853) — @louis.
  • returning to a coding chat keeps your chosen reasoning effort. starting a new chat no longer briefly replaces the conversation with a loading screen (#1860) — @louis.
  • linear workflow connections now refresh automatically. older connections may need one reconnect before teams and events work again (#1841) — @louis.
  • render tools can connect and run again after a connection error prevented them from loading (#1869) — @louis.
  • sandbox preview keeps looking for your server after an unavailable port. if the server is outside the workspace, preview explains where to start it (#1831) — @louis.
  • attachment cards stay visible and in order while files upload, and a resumed task's timer now measures the current run (#1846, #1848) — @louis.
1.27.02026-09-14
Added14 entries
  • start with a workflow template for a weekly shipping digest, daily inbox triage, or weekly revenue pulse. open workflow templates on the workflows page, review the saved draft, connect the tools it needs, and activate it when ready (#1845, #1850) — @louis.
  • continue a workflow from slack. connect the workspace bot in settings → channels → slack and ask your workflow to post in a public channel. anyone in that thread can reply to continue the same task, with its files and context, for up to 30 days. follow-ups use the workflow owner's connected tools and permissions (#1840, #1849) — @louis.
  • keep separate wikis for your company, a project, or a small group. give each its own instructions and share it with everyone or selected teammates. your existing pages stay in the company wiki, and old links still work (#1745, #1805) — @louis.
  • read, update, and add rows to google sheets through the google drive plugin. use enable sheets tools in the plugin's account settings to grant access, then refresh its tools (#1807) — @louis.
  • send email from chat through the gmail plugin. sending asks for approval by default; you can set its permission separately from labeling and other mail actions (#1826, #1839) — @louis.
  • connect doppler to use your project's secrets in cloud coding sessions. connect dash0 to investigate your application's logs, metrics, and traces from chat (#1834, #1843, #1830, #1832) — @louis.
  • start a workflow when jamie finishes a meeting, with its summary, attendees, and action items as context. set up the webhook in the jamie plugin's events section, then choose meeting completed as a workflow trigger. filter to internal meetings or meetings with outside guests (#1806) — @louis.
  • ask for a calculator, dashboard, timeline, or mockup as an interactive page beside your chat. revise, share, or download it like any other artifact. pages are self-contained and cannot access the network or your account (#1775) — @louis.
  • browse and edit your cloud sandbox files in the new files tab beside preview and terminal. save changes directly; if the agent changed the same file, review the conflict before overwriting it (#1821) — @louis.
  • choose a smaller or larger cloud sandbox for new coding sessions in settings → inference. workspace admins can compare sizes and running costs; existing sessions keep their current size (#1818) — @louis.
  • turn on the projects beta in preferences to group chats and tasks into folders in the sidebar. drag work into a project, collapse its folder, or start a new chat inside it (#1755, #1761, #1765, #1766, #1825) — @louis.
  • turn on the subagents beta in preferences to let opencompany split research among helpers and bring their findings back into one answer. expand a helper's row to follow its work. this uses more credits per message (#1749, #1756, #1764) — @louis.
  • turn on past session access in preferences to ask about your earlier private chats in this workspace, including archived ones. use it to review recent work or find an earlier decision (#1771) — @louis.
  • choose claude fable 5.1 for cloud claude code chats and workflow steps. availability through your subscription depends on your claude plan. see your remaining claude subscription allowance and reset times in settings → inference (#1767, #1780, #1750) — @louis.
Changed9 entries
  • tasks and workflows are now available to everyone, with no beta switch. ask chat to run a saved workflow by name when you want background work; chat no longer creates one-off tasks on its own. existing recurring routines still run (#1847, #1812) — @louis.
  • keep workflows personal or share them with your company. add several schedules or plugin events to the same workflow, and filter the list to find your own or shared work (#1817, #1822) — @louis.
  • choose which meetings or issues start a workflow. granola events can follow one folder and its subfolders; linear events can filter by team and status. event connections now live in each plugin's events section (#1793, #1772, #1803) — @louis.
  • set on, ask, or off for an individual plugin tool. allow routine labeling while keeping email sending or deletion behind approval. always allow in chat now applies only to the tool you approved (#1839) — @louis.
  • approval cards now show the action, affected files or command, and any quoted cost in one place. chats and tasks waiting for you keep a distinct sidebar indicator until you answer (#1774, #1827) — @louis.
  • find chat models and coding agents in separate tabs in the model picker. auto comes first, models covered by your shared subscription show included, and a sandbox indicator appears when you select a coding agent (#1828, #1835, #1837, #1842) — @louis.
  • open new chat from the sidebar to start work. manage recurring routines on the tasks page, open skills from the sidebar footer, and collapse recents when you want more space (#1819, #1813, #1751) — @louis.
  • type a url or path into the sandbox preview's address bar to open another page. the preview opens when your dev server starts, and the port picker appears when several servers are running (#1809) — @louis.
  • feedback sent from a chat or task now includes a link to that work, so you do not have to find and paste it yourself (#1808) — @louis.
Fixed8 entries
  • scheduled workflows run again after a bug stopped due schedules from being picked up (#1852) — @louis.
  • draft your next reply while a task is working. refreshing an active chat now keeps its timer and stop control visible, and a resumed turn counts its own running time (#1802, #1815, #1795) — @louis.
  • file uploads no longer leave tasks hanging, and publishing a revised artifact no longer fails because an earlier revision is empty (#1748, #1798) — @louis.
  • gemini image generation works again instead of failing before producing an image (#1754) — @louis.
  • coding sessions recover more reliably from an unresponsive sandbox and retain recovered output. claude code tasks can resume after reaching a turn limit, and choosing fable 5.1 now starts the selected model correctly (#1777, #1779, #1782, #1742, #1780) — @louis.
  • recent chats and tasks no longer disappear from the sidebar because of a list limit, and beta preference changes appear immediately (#1763, #1762) — @louis.
  • wiki source references now have readable labels, and granola references link to the original note (#1844) — @louis.
  • long granola meetings can now start their workflows even when the full transcript is too large to fetch (#1803) — @louis.
Removed1 entry
  • automatic imports from connected providers into wiki have been retired. existing pages remain; use workflows with plugin events for new updates. the workspace slack bot now continues workflow threads instead of answering standalone mentions or direct messages (#1772, #1840) — @louis.
1.26.02026-09-11
Added6 entries
  • turn on the for review beta in preferences to read chat replies and task results in one place. reply, approve actions, or continue a task without leaving the queue. recent read items stay in view, with an archive button to clear finished work (#1712, #1714, #1725) — @louis.
  • start new work from the for review reading pane on desktop when no conversation is open. send a prompt, attach a file, or run a workflow while keeping your place in the queue (#1729) — @louis.
  • start a workflow when granola finishes your meeting notes, with the meeting summary and attendees as context. connect granola with an api key in wiki sources, enable meeting notes ready in the plugin, then choose it as your workflow's event (#1741) — @louis.
  • update an official plugin in one click when update available appears. your connected accounts and event choices stay in place; review tool permissions again for the new version (#1721) — @louis.
  • publish posts, replies, polls, and images with alt text through the x plugin, then check how your posts performed. reconnect your x account if prompted to enable image uploads (#1706) — @louis.
  • choose deepseek v4 flash from the chat model picker (#1708) — @louis.
Changed3 entries
  • chats and tasks now share the sidebar, with unfinished work first. open a task to read its result and clear its unread dot (#1733) — @louis.
  • archive a row in the sidebar or for review queue and it disappears immediately. if the save fails, the row returns with an error message (#1728, #1731, #1733) — @louis.
  • selecting a workflow in the composer now shows its saved models. change a step's model or reasoning for this run without changing the workflow, and attach images or pdfs when its models support them (#1702) — @louis.
Fixed9 entries
  • reopening a paused task now shows its approval controls immediately. the complete request loads before you can approve it (#1704) — @louis.
  • linear teams now load for workflow events even with legacy brain disabled. if linear access is revoked, the editor links you to reconnect your account (#1722, #1735, #1740) — @louis.
  • skills remain available across chat turns, and codex reliably receives the skills selected for a chat or workflow. claude chats and tasks no longer fail before replying because of the skill-editing tool (#1719, #1720, #1716) — @louis.
  • claude's follow-up questions now keep suggested answers and your own answer in one control, and stay open when you send another message (#1711) — @louis.
  • codex tasks retry temporary startup timeouts, and github access in coding sessions no longer breaks when another connection refreshes its credentials (#1723, #1710) — @louis.
  • command-palette search opens faster with a large task history, while older archived work stays searchable (#1724) — @louis.
  • browsing plugins and workflows no longer uses up the request allowance needed to load connected sources or browser live views (#1730) — @louis.
  • convex setup now checks the key's required permissions before connecting and explains which permissions are missing (#1732) — @louis.
  • background tasks can no longer keep handing the same work to new tasks. start tasks, workflows, and recurring schedules from a main chat (#1705) — @louis.
1.25.02026-09-10
Added9 entries
  • connect your own hosted mcp server from plugins to use tools beyond the catalog. test the connection, review its tools, and choose which actions need your approval (#1655) — @louis.
  • notion, supabase, resend, and convex are now official plugins. connect your account or deployment to work with documents, databases, and email from chat (#1640, #1651, #1659, #1669) — @louis.
  • google workspace admins can now use the google admin plugin to create user accounts, manage groups, and add group members (#1686) — @louis.
  • keep skills personal or share them with your company through the visibility setting. company skills are available for teammates to use and edit (#1668) — @louis.
  • create, edit, rename, or archive your skills from chat, including codex and claude code sessions. you can also edit instructions directly in the skill's settings page (#1646, #1685) — @louis.
  • turn on bots in preferences to create a named bot for ongoing work and return to its conversation from the sidebar (#1661, #1665) — @louis.
  • choose claude opus 5 for claude code chats and workflow steps (#1663) — @louis.
  • see your remaining codex subscription allowance and when it resets in inference settings (#1683) — @louis.
  • reschedule a google calendar event from chat while keeping its attendees and meeting link (#1670) — @louis.
Changed6 entries
  • plugins now belong to you personally, with your own accounts and permissions. if a previously installed plugin is missing, reinstall it from plugins, review its permissions, and reconnect your account when prompted (#1679, #1689) — @louis.
  • coding sandbox time now uses workspace credits and appears as sandbox usage, even with a connected model subscription. running time includes idle time; paused time is not charged (#1681) — @louis.
  • connect stripe through its sign-in and consent flow without creating and pasting an api key (#1673) — @louis.
  • unfinished workflows now save as drafts. add instructions to every step, then activate the workflow when you're ready (#1658) — @louis.
  • background tasks now pause for one-time action approval across all engines. approve or deny from the task, then it resumes without changing your standing permissions (#1654, #1660) — @louis.
  • requesting a demo now starts with a short form about your team, then lets you choose a time on the same page (#1703) — @louis.
Fixed14 entries
  • unsent message text and mentions now survive a refresh in the same browser tab. attachments still need to be added again (#1684) — @louis.
  • skills selected in workflow instructions now keep their readable names and work after saving, fixing runs that incorrectly reported a skill as unavailable (#1696, #1697) — @louis.
  • plugin installs are more reliable, and returning to the catalog after installing now shows the correct state (#1645, #1647, #1692) — @louis.
  • plugins that still need an account now show a connection warning and a clear connect button (#1680) — @louis.
  • choosing always allow now clears all pending requests for that same action in the current chat run, so work can continue (#1699) — @louis.
  • connected accounts load reliably again, including after connecting notion (#1596, #1609, #1650) — @louis.
  • approving an action now resumes the original work without restarting it or asking again. repeated identical writes within a turn no longer create duplicate results (#1631, #1670) — @louis.
  • tasks waiting for approval stop showing a working timer, and tasks waiting for your input can now be archived (#1676, #1644) — @louis.
  • coding sessions recover when a sandbox stops responding, and claude tasks reliably resume at their scheduled follow-up time (#1672, #1678, #1635, #1694) — @louis.
  • replacing a google doc no longer turns headings and ordinary paragraphs into list items (#1638) — @louis.
  • opening one tool call in an older message no longer expands the others (#1690) — @louis.
  • command-palette search keeps matching chats and tasks in newest-first order, including after you clear the search (#1698) — @louis.
  • connected tools can recover from an invalid input within the same turn, so you no longer need to send another message just to retry a corrected request (#1691) — @louis.
  • disabling legacy brain now stops its background imports and related usage charges (#1641) — @louis.
Security1 entry
  • updated dependencies to address reported security vulnerabilities (#1639) — @louis.
1.24.02026-09-07
Added4 entries
  • Chat can now put a document or draft it writes into a side panel — resize it, open it full-screen, and come back to it later — instead of leaving it buried in the conversation (#1600, #1606) — @louis.
  • Workflows can now be triggered by plugin events: turn on the events you care about, per workspace, and the right workflow kicks off on its own — starting with a new Linear issue (#1602) — @louis.
  • The Gmail plugin can now download an email's attachments — pull the original file into a coding session or hand it straight to you, instead of only reading the message (#1622) — @louis.
  • Watch a connected browser profile as it works: authenticated browser sessions now show a live view right in chat while opencompany researches a page that needs you signed in (#1601) — @louis.
Changed4 entries
  • New Codex coding sessions now default to GPT-6 Astra, with Sol, Terra, and Luna still one pick away (#1627) — @louis.
  • Infisical is now an official plugin: connect it once from its plugin page and choose exactly what it can do, instead of the old Integrations card (#1595, #1598) — @louis.
  • Provider connections now live entirely in Plugins — the separate Integrations settings page is gone, and your authenticated browser profiles moved to Account settings (#1607) — @louis.
  • Filter the plugin catalog down to just what you've installed, so your connected tools are one click away (#1604) — @louis.
Fixed6 entries
  • Google Docs rewritten from Markdown now keep headings, lists, checkboxes, and inline formatting instead of displaying raw Markdown as plain text (#1594) — @louis.
  • Reasoning you've expanded in a chat now stays open as the conversation updates, instead of collapsing on you (#1593) — @louis.
  • Home now shows all your active tasks instead of quietly hiding some of them (#1605) — @louis.
  • Coding sessions keep their GitHub access through a full turn, fixing a case where authorization could drop mid-run (#1599) — @louis.
  • Tasks that are waiting on you now hold their "waiting" status instead of appearing to reset (#1626) — @louis.
  • GPT-6 Astra Codex sessions now start successfully instead of failing before the first response because the sandbox runtime did not yet recognize the model (#1628) — @louis.
1.23.02026-09-04
Added7 entries
  • A wave of new official plugins: connect Gmail, Google Drive, HubSpot, Stripe, PostHog, Attio, Fathom, and Latitude, then pick exactly which actions opencompany is allowed to take and approve anything sensitive before it runs (#1515, #1517, #1564, #1562, #1557, #1583, #1584, #1585) — @louis.
  • Edit your Google Docs from chat: the Google Drive plugin can now make changes to an existing doc, not just read it, with each edit gated behind your approval (#1573) — @louis.
  • GPT-6 Astra is available across chat and Codex, and runs against your own connected ChatGPT/Codex subscription at no extra cost when you have one (#1566) — @louis.
  • Tasks now have an activity feed: watch each run's steps as they happen, retry or reopen a task, and pick it back up just by leaving a comment (#1542) — @louis.
  • Filter tasks by recent time ranges — including the last 24 hours and last 2 days — and your choice now follows you across navigation and devices (#1571, #1581) — @louis.
  • Jump to any task from the command palette: active and archived tasks and chats show newest-first and are searchable by name, prompt, or ID (#1554) — @louis.
  • Hover the model icon in a session header to see which model that session is running (#1559) — @louis.
Changed5 entries
  • Plugins now have their own spot in the left sidebar, and the catalog is redesigned around search, featured picks, and categories so you can find the right plugin faster (#1569, #1565) — @louis.
  • Granola, Jamie, and X now connect through official plugins instead of built-in integrations. Reconnect each one once through its plugin to keep it working (#1582, #1588, #1556) — @louis.
  • Your personal Codex and Claude Code subscriptions have moved from Integrations to the new Inference settings, kept separate from workspace-wide model access (#1567) — @louis.
  • Long chats and coding sessions stay fast: busy conversations now compact their oldest turns behind the scenes while keeping the current turn and recent context intact (#1563) — @louis.
  • Chat attachment uploads are more reliable — an interrupted upload resumes or replays cleanly instead of failing (#1502) — @andrei.
Fixed8 entries
  • Attachments work in task conversations again: drop a screenshot or file into a task follow-up just like in chat (#1539) — @louis.
  • Long, tool-heavy chats read cleanly again — messages, thinking, and tool calls stay in order while a task runs, then collapse behind a single disclosure once it's done (#1561, #1549) — @louis.
  • The context indicator now shows how full a chat's context actually is, instead of inflating after multi-step tool runs (#1558) — @louis.
  • Skill search is more forgiving: a relevant skill no longer disappears just because your query included an extra word (#1560) — @louis.
  • Tasks no longer get stuck retrying forever — a run that can't finish cleanly is now stopped instead of restarting indefinitely (#1551, #1552) — @louis.
  • Tasks recover from transient coding-engine setup failures, and can repair a missing final response instead of marking otherwise completed work as failed (#1579, #1587) — @louis.
  • Claude coding sessions start reliably again, fixing authorization and readiness checks that could fail a session before your prompt even ran (#1553, #1550) — @louis.
  • Connected actions like PostHog stay available across turns instead of needing to be rediscovered mid-conversation (#1568) — @louis.
Removed2 entries
  • The legacy workspace GitHub integration is retired — repository access and coding sessions now run through the personal GitHub plugin (#1589) — @louis.
  • The iMessage integration has been removed (#1580) — @louis.
Security1 entry
  • Upgraded dependencies to clear reported high-priority vulnerabilities (#1578) — @louis.
1.22.02026-09-03
Added6 entries
  • Official plugins let you connect your own tools and pick exactly which actions opencompany can take: Slack, GitHub, Google Calendar, Render, SigNoz, and Better Stack are all available now (#1505, #1494, #1513, #1541, #1533, #1496) — @louis and @opencompany-bot.
  • Add a plugin in one click, straight from the plugin overview, instead of stepping through a separate review first (#1536) — @louis.
  • Skills-only plugins are here: install a playbook like yc-advise that adds its own / commands, no connected tool required (#1525) — @louis.
  • Connect your personal GitHub account, kept separate from workspace ingestion, so coding sessions and tools act as you — and opencompany now walks you through granting repository access when it's missing (#1475, #1506) — @louis.
  • Route chat and task turns through your own ChatGPT/Codex subscription: pick a connected account in the new Inference settings and GPT 5.6 Sol and Terra run against it at no extra cost (#1488, #1493) — @opencompany-bot.
  • Ask opencompany to update one of your workspace skills right from chat (#1519) — @louis.
Changed6 entries
  • Wiki is now your workspace's knowledge system by default. Owner onboarding is shorter, and connecting sources or importing company context stays available from Wiki after setup. Brain stays available behind a legacy setting (#1527, #1543, #1548) — @louis.
  • The old built-in Slack integration is retired in favor of the Slack plugin; reconnect Slack once through the plugin to keep it working (#1508) — @louis.
  • The workspace Usage page is rebuilt around one daily-spend chart, broken down by where your spend goes, so you can see how much you're spending and which way it's trending at a glance (#1512) — @louis.
  • Long coding chats stay readable: earlier Codex and Claude tool calls collapse behind a single disclosure while the latest reply and anything still running stay in view (#1529) — @louis.
  • Long chats load faster — older reasoning and tool details now load only when you expand them (#1487) — @opencompany-bot.
  • Reaching your Hobby plan's member limit now explains what's going on and links straight to the upgrade, instead of a silently disabled invite form (#1530) — @louis.
Fixed8 entries
  • Pending action approvals show up right away in older chats, with their full details and controls, instead of waiting on a separate fetch (#1522) — @louis.
  • Reloading a Claude or Codex chat keeps the model you were using — Fable no longer quietly switches to another model on your next message (#1486) — @opencompany-bot.
  • The built-in opencompany connector no longer occasionally shows as disconnected at the start of a Claude session (#1544) — @louis.
  • Workflow tasks get a proper generated title instead of keeping the raw workflow name (#1524, #1538) — @louis.
  • Follow-up questions on report-producing tasks now answer in the conversation instead of creating another report when you ask for a summary or clarification (#1540) — @louis.
  • Scheduled follow-ups from Claude tasks now run at the requested time instead of appearing scheduled and then never starting (#1546) — @louis.
  • Wiki commands no longer fail when GPT 5.6 leaves an optional input blank, so tasks can keep browsing and updating pages (#1547) — @louis.
  • Messages with emoji or other unusual characters save reliably instead of sometimes failing a reply midway (#1511) — @louis.
1.21.02026-09-01
Added2 entries
  • The official Linear plugin now installs in one click: pin it to your workspace, choose exactly which Linear actions opencompany is allowed to take, and approve anything sensitive before it runs — replacing the old Linear integration card (#1468) — @louis.
  • Coding sessions can now use your wiki: Codex and Claude sandboxes reach the same wiki tools as chat, so a coding task can pull context from your pages or write results back (#1480) — @opencompany-bot.
Changed1 entry
  • Long, busy chats stay fast: heavy conversations now refresh their snapshot behind the scenes instead of slowing down as they grow (#1485) — @opencompany-bot.
1.20.02026-08-28
Fixed2 entries
  • Coding sessions hold on through more dropped connections: when a long-running command loses its stream mid-run, the session now reconnects and keeps going instead of failing (#1459) — @louis.
  • Scheduled workflows you set up before Skills launched now run again on schedule, instead of failing to start (#1457) — @opencompany-bot.
1.19.02026-08-27
Added7 entries
  • Skills and Plugins are here: install reusable skills and plugins into your workspace so opencompany follows your own playbooks and reaches more tools, each one sandboxed and version-pinned so an update never quietly changes what a skill does (#1416) — @louis.
  • Run a skill in chat by typing / and picking it, the same way you'd reach for a slash command (#1420) — @opencompany-bot.
  • Write your own skill without leaving opencompany: a new "New skill" flow in Settings takes a name, description, and instructions, and gives it its own / command (#1436) — @louis.
  • Turn a good conversation into a reusable skill straight from chat — just ask, and opencompany saves it for your workspace (#1449) — @opencompany-bot.
  • Your wiki can now build itself from the tools you already use: connect Gmail, Slack, Granola, Jamie, Linear, or GitHub, choose what's in scope, and a librarian keeps the right pages written and organized for you (#1419) — @louis.
  • Filter the Tasks board and list by workflow, so you can focus on the runs from one workflow at a time (#1452) — @opencompany-bot.
  • A Docs link now sits in the account menu, putting the documentation one click away (#1441) — @opencompany-bot.
Fixed6 entries
  • Signing in reliably opens your workspace again for accounts tied to older organizations, instead of dead-ending on a "We couldn't open this page" error after you pick a workspace (#1415) — @louis.
  • Launch several background chats back to back: the composer clears and refocuses the moment each one is accepted, instead of waiting on the last one to finish (#1424) — @opencompany-bot.
  • A brief network blip while a reply is streaming no longer surfaces an error — the chat reconnects and picks up where it left off (#1425) — @opencompany-bot.
  • Pinned and recent chats in the sidebar now appear immediately on load instead of shifting into place a moment later (#1427) — @opencompany-bot.
  • Coding sessions are steadier: they recover when a command stream drops mid-run, and tool calls keep showing their real input and output instead of a blank placeholder (#1445, #1417) — @opencompany-bot.
  • Starting a new chat right as we ship an update no longer occasionally fails — your first message goes through (#1421) — @louis.
1.18.02026-08-24
Added2 entries
  • AI image generation is now available as a chat capability, so you can create or restyle images from a prompt and an optional reference image, right from chat (#1375) — @opencompany-bot.
  • Grok 4.6 is now available as a chat model (#1392) — @opencompany-bot.
Changed2 entries
  • The macOS desktop app has a redesigned title bar that shows what you're looking at and gives quick access to a new chat, the sidebar, and settings (#1384) — @louis.
  • Recent chats on Home now stay visible for seven days instead of just one (#1383) — @opencompany-bot.
Fixed5 entries
  • Coding sessions now recover automatically from interrupted deploys and dropped command streams instead of getting stuck, and long coding chats load faster (#1370, #1371, #1372, #1373, #1390) — @louis and @opencompany-bot.
  • Terminal and Preview access in coding sandboxes works again for sessions started before a recent update (#1388) — @opencompany-bot.
  • Approving a tool action in an older chat no longer breaks the conversation view (#1386) — @opencompany-bot.
  • Long wiki page link titles are now truncated instead of overflowing (#1389) — @opencompany-bot.
  • Signing in now reliably syncs your account instead of occasionally failing right after login (#1374) — @louis.
1.17.02026-08-21
Added2 entries
  • opencompany can now trigger a workflow automatically when a Linear issue enters triage, so the right task kicks off the moment something needs attention (#1347) — @opencompany-bot.
  • Coding sandboxes now include Docker, so tasks that need to build or run containers can do that directly inside a coding session (#1336) — @opencompany-bot.
Changed3 entries
  • Live coding chat now shows clear, labeled actions — reads, writes, edits, searches, commands, and plans — with file names and a preview of the assistant's thinking, instead of raw tool output (#1359, #1361) — @opencompany-bot.
  • Coding sessions can now run for up to three hours before timing out, up from one hour, giving longer tasks room to finish (#1363) — @opencompany-bot.
  • Wiki folders now remember which ones you had open the last time you visited (#1356) — @opencompany-bot.
Fixed3 entries
  • Opening an active task no longer hides your earlier messages, and its status no longer gets stuck on "Connecting" while work is still happening (#1358) — @opencompany-bot.
  • A coding session that loses its connection now recovers reliably instead of getting marked as failed while it's still working in the background (#1360) — @opencompany-bot.
  • Listing comments on a LinkedIn post works again (#1350) — @opencompany-bot.
1.16.02026-08-20
Added5 entries
  • Run a scheduled workflow right away with a new "Run now" action on its editor page, instead of waiting for its next scheduled time (#1346) — @opencompany-bot.
  • Choose which model handles a task you spawn from chat, instead of always using the workspace default (#1342) — @opencompany-bot.
  • opencompany now has a native macOS desktop app, with secure Google sign-in handoff from your browser (#1328, #1331) — @louis.
  • opencompany can now read Gmail attachments and Google Drive files, like PDFs and spreadsheets, when helping you (#1335) — @opencompany-bot.
  • Wiki pages can now be organized into folders, with a smoother editor and reliable page-link suggestions (#1325) — @louis.
Changed2 entries
  • If a chat reply gets cut off partway through, you now keep everything it already wrote along with a clear note about what stopped, instead of losing the response (#1332) — @louis.
  • Clicking a chat in the sidebar now opens it instantly instead of after a noticeable delay (#1290) — @opencompany-bot.
Fixed15 entries
  • Chats and coding sessions are more resilient during app updates: they recover automatically instead of getting stuck or breaking with an error mid-conversation when we ship a routine update (#1338, #1314) — @louis.
  • Reopening a task no longer briefly shows it as empty while your message history reloads (#1341) — @louis.
  • Expanding a tool call in chat now shows exactly what was sent and returned, instead of a generic placeholder (#1343) — @louis.
  • Sending a follow-up message to a finished task now works reliably instead of sometimes failing (#1339) — @louis.
  • Invoking a workflow by typing "#workflow-name" in chat now actually works (#1333) — @louis.
  • Starting a background chat with "&" now correctly uses your default model instead of copying whatever session you happened to be viewing (#1322) — @louis.
  • Finished chat replies no longer keep showing a "still running" timer after they're done (#1319) — @louis.
  • Claude Code sessions no longer sometimes mislabel themselves as Codex in the status indicator (#1312) — @opencompany-bot.
  • Editing a workflow, task, recurring task, or wiki page no longer shows a false "save failed" message right after it actually saved (#1296, #1291) — @louis.
  • A brand-new chat started from Home no longer gets stuck on a broken page if you click away too quickly (#1288) — @louis.
  • The sidebar's unread and "working" indicators for chats are now more accurate (#1292) — @louis.
  • A background chat you sent is no longer shown as failed just because the sidebar took a moment to catch up (#1286) — @louis.
  • Coding chats no longer get stuck endlessly retrying, or stuck on "Connecting" with no messages (#1283) — @louis.
  • Looking up a lead's email that genuinely has none on file no longer shows a failed action and retries (#1285) — @louis.
  • Stripe billing webhook processing, which keeps your subscription and payment status in sync, no longer silently fails (#1282) — @louis.
1.15.02026-08-17
Fixed1 entry
  • If your connection drops or you reload the page while a tool call, like a browser action, is still running, the reply now reliably picks back up instead of risking a broken or duplicated result (#1269) — @louis.
1.14.02026-08-14
Changed1 entry
  • The marketing homepage has a refreshed hero message and simpler navigation (#1247) — @louis.
Fixed6 entries
  • Starting a new Codex coding session no longer occasionally leaves you on a broken chat URL if you navigate away before the message finishes saving (#1245) — @louis.
  • Claude Code and Codex sessions now reliably keep their full conversation history instead of occasionally losing earlier context (#1259) — @louis.
  • Chat reasoning no longer gets dropped while a reply is still streaming in (#1254) — @louis.
  • If your browser tab is still running an older version after we ship an update, you now get a clear prompt to refresh instead of chat silently breaking (#1249) — @louis.
  • Content saved to Brain now reliably kicks off indexing right away instead of occasionally waiting for the next sync (#1215) — @opencompany-bot.
  • The chat title now lines up with the left edge of the header above it (#1255) — @louis.
1.13.02026-08-12
Fixed1 entry
  • Approving a tool call, like a browser action, no longer occasionally gets silently dropped and marks the task as finished before it actually is (#1193) — @opencompany-bot.
Removed1 entry
  • opencompany Quick, the macOS menu bar prototype (#876, #877), is no longer available (#1195) — @opencompany-bot.
1.12.02026-08-11
Changed1 entry
  • Chat runs on a rebuilt real-time backend: replies stream in faster and pick up right where they left off if your connection drops or you reload the page mid-response (#1167, #1168, #1169, #1170, #1172, #1173, #1175, #1176, #1177, #1178, #1179, #1180, #1181, #1182) — @louis and @opencompany-bot.
Fixed1 entry
  • Tool call results now stay in the right order instead of jumping around while a reply is still streaming in (#1183) — @louis.
1.11.02026-08-10
Added2 entries
  • Import skills from a public GitHub repo or a skills.sh page by pasting a link, so your workspace can reuse skills other people have published instead of writing your own from scratch (#1140) — @opencompany-bot.
  • Qwen 3.8 Max is now available as a chat model (#1153) — @opencompany-bot.
Changed2 entries
  • Wiki pages are faster and closer to Notion: edits show up instantly, and a new / slash menu lets you insert blocks and links without leaving the keyboard (#1148, #1149) — @opencompany-bot.
  • Cmd+K now shows archived chats alongside active ones, sorted by recency, so you can jump back into an old conversation without digging through the archive separately (#1147) — @opencompany-bot.
Fixed6 entries
  • Chats that stalled mid-response now recover on their own instead of hanging indefinitely (#1146) — @opencompany-bot.
  • Archiving a chat no longer risks crashing the page — you'll see a clear error and can retry if something goes wrong (#1155) — @opencompany-bot.
  • A chat's unread status and a task's ability to keep going no longer break right after we ship an update (#1150, #1151, #1154) — @opencompany-bot.
  • If a model provider errors out mid-reply, you keep whatever it had already written instead of losing the response (#1152) — @opencompany-bot.
  • Shared links to the pricing, use-cases, media, ship, and blog pages now show a preview image on X, WhatsApp, and Slack (#1126) — @opencompany-bot.
  • The chat header now lines up correctly with the sidebar above it (#1134) — @opencompany-bot.
1.10.02026-08-07
Added4 entries
  • Post to X (Twitter) right from chat — connect one or more accounts in Settings, then ask opencompany to post to one account or several at once (#1109, #1121) — @opencompany-bot.
  • opencompany can now generate real files in chat — spreadsheets, PDFs, slides, docs, and images — that you can preview, download, and share, instead of only markdown replies (#1137) — @opencompany-bot.
  • Wiki (preview): a self-building company wiki that lives alongside chat, with linked pages, cited sources, and a Notion-style editor. Turn it on under Preferences > Beta features (#1133, #1136, #1138, #1139, #1143) — @opencompany-bot.
  • A new pricing page lays out the Hobby and Pro plans side by side, and the marketing site nav gained a "Book demo" option (#1110) — @opencompany-bot.
Changed4 entries
  • Onboarding is shorter — new workspaces get their Brain folders set up automatically instead of asking you to arrange them upfront (#1114) — @opencompany-bot.
  • Chats you start now show up in the sidebar the moment you send them, instead of a moment later (#1130, #1135) — @opencompany-bot.
  • The changelog moved from the sidebar footer into the account menu (#1115) — @opencompany-bot.
  • Popovers, dialogs, and menus now use softer, more natural shadows instead of hard-edged ones (#1124) — @opencompany-bot.
Fixed10 entries
  • Chat tables no longer split words letter-by-letter into narrow columns (#1128) — @opencompany-bot.
  • A chat open in two browser tabs now shows the same running or finished status in both, instead of disagreeing (#1129) — @opencompany-bot.
  • Approving an X post now shows a clear, readable summary of what's about to be posted instead of raw data (#1125) — @opencompany-bot.
  • Approving an action no longer briefly flashes a false "could not produce a response" error before the real answer (#1118) — @opencompany-bot.
  • Claude Code no longer shows your account as disconnected just because you temporarily hit a usage limit (#1127) — @opencompany-bot.
  • Claude Code skills are discovered reliably again, and skills mentioned when starting a workflow are no longer dropped (#1131, #1120) — @opencompany-bot.
  • Claude Code sessions that run background subagents now reliably return with an answer instead of stopping silently (#1113) — @opencompany-bot.
  • Coding tasks recover automatically after brief sandbox network hiccups instead of failing outright, and archiving a chat now reliably stops any coding work still scheduled for it (#1112, #1111) — @opencompany-bot.
  • Codex sessions stay connected instead of periodically asking you to reconnect (#1117) — @opencompany-bot.
  • Connecting Infisical is more reliable, including for the EU region and for sandboxes that previously failed to provision secrets (#1116, #1119, #1122) — @opencompany-bot.
1.9.02026-08-06
Added3 entries
  • opencompany Hobby: a free plan with $5 of included usage refreshed monthly, for single-member workspaces, alongside Pro (#1095) — @opencompany-bot.
  • Connect Infisical as a workspace integration in Settings so Codex and Claude Code coding sandboxes can pull your secrets automatically (#1096) — @opencompany-bot.
  • Chat and Brain now accept CSV, TSV, Markdown, plain text, and JSON uploads, not just images and PDFs (#1104) — @opencompany-bot.
Changed2 entries
  • Cmd+K is now a real command palette: search your chats and jump to actions from one input, with compose as a separate step instead of crowding the search box (#1093) — @opencompany-bot.
  • Signing in and signing up now happen on our own pages instead of a WorkOS hosted picker — pick Google or email magic link, and we remember which one you used last (#1099) — @opencompany-bot.
Fixed2 entries
  • The opencompany home page's chat and task list scrolls properly again instead of getting cut off (#1098) — @opencompany-bot.
  • Fixed a bug where interrupting a tool call, like a browser action, could break the next message you sent in that chat (#1103) — @opencompany-bot.
1.8.02026-08-05
Added3 entries
  • opencompany can now create Attio records — people, companies, and deals — directly from chat, alongside the existing search, update, and comment support (#1068) — @opencompany-bot.
  • @claude now starts a Claude Code session from the composer, the same way @codex already does (#1071, #1091) — @opencompany-bot.
  • Composer drafts now save per chat — switch away mid-sentence and your unsent text is still there when you come back (#1069) — @opencompany-bot.
Changed3 entries
  • Background chats started with & can now be launched from any chat, including while another chat or Codex session is still running, and the directive is highlighted as you type it (#1064, #1070, #1081, #1088) — @opencompany-bot.
  • opencompany usage is now billed at cost with no platform markup, and included monthly usage is tracked separately from top-up funds (#1083) — @opencompany-bot.
  • Typing in the coding workspace terminal now feels instant instead of waiting on the remote echo (#1062, #1076) — @opencompany-bot.
Fixed9 entries
  • The sidebar no longer shows a background chat as unread before it has actually started running (#1086) — @opencompany-bot.
  • The sidebar's working indicator no longer lags behind a chat that keeps generating after you navigate away (#1065) — @opencompany-bot.
  • Task sessions no longer show duplicate task cards, and @skill mentions work again inside them (#1075, #1092) — @opencompany-bot.
  • Starting a Codex task from Kimi, Grok, or Claude in main chat now hands off to the right model instead of misrouting it (#1079) — @opencompany-bot.
  • Fixed a race that could mark a Codex coding task as failed even though it was still running (#1072) — @opencompany-bot.
  • The coding workspace preview picker correctly detects open ports again (#1090) — @opencompany-bot.
  • The chat composer no longer overlaps the latest message when it grows tall (#1085) — @opencompany-bot.
  • Opening a chat from the sidebar now focuses the composer automatically (#1078) — @opencompany-bot.
  • Automatic model routing is now more reliable, with better diagnostics when something goes wrong (#1084) — @opencompany-bot.
1.7.02026-08-04
Added5 entries
  • opencompany Pro: a $20/month plan for a single workspace with room for up to 10 members, managed through Stripe billing (#1052) — @opencompany-bot.
  • Voice dictation in the composer — tap the mic, watch your words appear live, and edit before sending (#1054) — @opencompany-bot.
  • Connect Neon as a read-only integration to ask about your database's projects, branches, schemas, and data (#1056, #1059) — @opencompany-bot.
  • opencompany can now post comments on Attio records, list entries, and threads (#1055) — @opencompany-bot.
  • Start a background chat straight from the composer by beginning your message with & (#1060) — @opencompany-bot.
Changed1 entry
  • Integrations that lose their connection now show a clear "Needs reconnect" state in Settings instead of failing silently (#1053) — @opencompany-bot.
Fixed2 entries
  • The sidebar's working indicator no longer lags behind chats that are actively running (#1051) — @opencompany-bot.
  • The coding workspace preview picker no longer lists internal sandbox ports as if they were your app (#1058) — @opencompany-bot.
1.6.02026-08-03
Added3 entries
  • Read, update, and append rows in Google Sheets from chat, alongside the existing Google Docs support (#1047) — @opencompany-bot.
  • Share links now cover task runs and Codex/Claude Code cloud coding sessions too, not just regular chats (#1043, #1044) — @opencompany-bot.
  • Finished tasks that resulted in a GitHub pull request now show a direct link to it in the task's Properties panel (#1015) — @opencompany-bot.
Changed2 entries
  • The sidebar and chat list status indicator now shows a distinct spinner for chats actively running, instead of only marking finished ones unread (#1046) — @opencompany-bot.
  • Workflow steps now hand off to the next step with a focused brief instead of the whole prior transcript, and each step's activity shows up in the task's timeline (#1036, #1048) — @opencompany-bot.
Fixed6 entries
  • Shift+Enter in the composer now inserts a newline instead of sending the message (#1035) — @opencompany-bot.
  • Chats no longer show a stale unread indicator after a live reply has already arrived (#1042) — @opencompany-bot.
  • Scheduled tasks and workflows now fire reliably instead of occasionally being skipped (#1039) — @opencompany-bot.
  • Interrupted chat turns now recover cleanly, without losing the assistant's reply or double-sending an iMessage notification (#1040) — @opencompany-bot.
  • Looking up a company's employees now returns clean results, with a clear prompt instead of an error when nothing matches (#1049) — @opencompany-bot.
  • Fixed a database migration gap that could silently block iMessage notifications from sending (#1045) — @louis.
1.5.02026-08-01
Added11 entries
  • Get notified over iMessage when opencompany needs your input or finishes a task, with a secure device-pairing flow (#1025) — @louis.
  • Connect an authenticated browser profile so opencompany can research pages that need you to be logged in; sensitive sites like banking are blocked by default (#1028) — @opencompany-bot.
  • Workflows can now run on a schedule, set with a plain-language frequency builder — every N minutes or hours, daily, weekdays, or weekly — instead of raw cron (#1009, #1020) — @opencompany-bot.
  • Start a one-off background task straight from the composer with #task, no need to save it as a workflow first (#987) — @opencompany-bot.
  • Pick the coding model for each step in a workflow, and use Claude Code as a workflow step engine (#1005, #1022) — @opencompany-bot.
  • Workflow tasks started from chat can now carry attachments (#1012) — @opencompany-bot.
  • Tasks and workflows are now visible to your whole workspace, not just whoever started them (#1013) — @opencompany-bot.
  • The Tasks page can switch to a dense list view alongside the existing board (#992) — @opencompany-bot.
  • Claude Code sessions can now read from your connected tools — Slack, Gmail, Linear, and more — matching what Codex sessions could already do (#989, #1008) — @opencompany-bot.
  • The sidebar now shows a live status dot on each chat for running, waiting-on-you, and errored states (#1011) — @opencompany-bot.
  • Codex chats show their live plan as a running checklist while they work (#1010) — @opencompany-bot.
Changed3 entries
  • @mentioning Claude Code or Codex in the composer now applies to that one message only, instead of switching the chat's engine for good (#1016) — @opencompany-bot.
  • New workflows are active by default instead of needing a manual toggle, and workflow mentions are now highlighted in the composer (#1002, #1007) — @opencompany-bot.
  • Workflow step lists now render in the read-only run detail view (#1030) — @opencompany-bot.
Fixed9 entries
  • Long Linear issue descriptions and comments up to Linear's own limit are no longer rejected early (#1032) — @opencompany-bot.
  • Stopping a task now reacts immediately across Claude Code, Codex, and opencompany chat engines (#1023) — @louis.
  • Scheduled task wakeups now reliably fire across all chat engines (#1024) — @louis.
  • The sidebar no longer flashes stale data on refresh (#1026) — @opencompany-bot.
  • Cloud coding workspace terminals no longer freeze while typing during a slow background operation (#982) — @louis.
  • The cloud terminal no longer offers to autofill saved passwords (#986) — @opencompany-bot.
  • Fixed a bash syntax bug that could block every cloud coding session from starting (#979) — @louis.
  • Tasks and workflow runs now hold up better through runner restarts and deploys, without losing chat history or duplicating messages (#988, #990, #991, #993, #995, #1019, #1031) — @opencompany-bot and @louis.
  • Assorted workflow composer fixes: mentions stay available in engine mode, the caret stays aligned, duplicate stop controls are gone, and task titles and sources display correctly (#994, #996, #997, #999, #1000, #1003, #1004, #1006) — @opencompany-bot.
1.4.02026-07-29
Added10 entries
  • New Tasks & Workflows board: a Kanban view of every chat and workflow run, with a time-range filter and a Linear-style detail panel where you can reply to a task like a normal chat (#957, #963, #964, #973, #947) — @louis and @opencompany-bot.
  • Workflows now run as sequential multi-step tasks with a model and skill per step, and can be started directly from main chat just by asking (#967, #953, #948) — @louis and @opencompany-bot.
  • Codex and Claude Code chats get a persistent Preview and Terminal workspace, tucked behind a single toggle in the chat header until you need it (#962, #972, #976) — @louis and @opencompany-bot.
  • Configure per-repository environment and setup for Codex and Claude Code sandboxes, stored encrypted (#956) — @louis.
  • Codex sessions can now save findings to the Brain, same as main chat (#965) — @opencompany-bot.
  • Ask opencompany to create a Google Doc, optionally pre-filled with text (#950) — @opencompany-bot.
  • Connect PostHog to ask about dashboards, saved insights, and analytics queries from chat (#969) — @opencompany-bot.
  • Claude Fable 5 and DeepSeek V4 Pro are now available as chat models (#975, #946) — @opencompany-bot.
  • Shared chat links now show the chat's title and a branded preview card when unfurled elsewhere (#961) — @opencompany-bot.
  • Automatic model routing (beta): turn it on under Preferences > Beta features to have opencompany pick the model for your first message (#966, #968, #970) — @opencompany-bot.
Changed1 entry
  • Cmd+K now opens the full composer — attachments, @mentions, and the model picker included — instead of a bare text box, and always starts a new chat in the background (#949) — @opencompany-bot.
Fixed4 entries
  • Claude Code sandboxes no longer race two agents on the same chat after a runner restart, which could revert files or duplicate commits (#974) — @louis.
  • Scheduled Claude Code wakeups now reliably fire even across restarts (#971) — @louis.
  • Removing a Linear issue's project from chat now actually clears it instead of being silently ignored (#958) — @louis.
  • The onboarding message that kicks off Brain building now always sends on the first try, and new chats default to Kimi K3 (#955) — @opencompany-bot.
Removed1 entry
  • Removed the retired Local Codex bridge, including its beta preference and composer option; leftover Local Codex chat sessions were cleaned up (#959, #960) — @opencompany-bot.
1.3.02026-07-28
Added10 entries
  • Claude Code is now a persistent sandbox chat engine in opencompany, with subscription authentication, model and effort controls, attachments, resumable sessions, and expandable subagent traces (#928, #930, #935) — @louis and @opencompany-bot.
  • Build durable workspace workflows and reusable skills, edit their instructions with rich Markdown, and mention skills from workflow instructions with autocomplete (#927, #934, #937) — @louis and @opencompany-bot.
  • Main chat can discover and use relevant workspace skills without requiring an explicit @skill mention (#909) — @opencompany-bot.
  • Cloud Codex tasks can safely read from connected tools that are enabled for read-only access, without exposing provider credentials to the sandbox (#914) — @opencompany-bot.
  • opencompany can create Gmail drafts without sending them, controlled by a separate per-account permission (#922) — @opencompany-bot.
  • Latitude is now available as a personal integration for chat and background tasks, with per-capability read and write controls (#925) — @opencompany-bot.
  • Switch between workspaces or create a new one directly from the workspace picker (#901) — @opencompany-bot.
  • Search within timestamped YouTube transcripts or bring a complete transcript into chat (#920, #933) — @louis and @opencompany-bot.
  • Look up a known prospect's work email from their name and company or LinkedIn profile, without re-running broad prospect research (#938) — @opencompany-bot.
  • Workspace admins can cap paid capability spend per chat, with one-off approval when a request would exceed the budget (#941) — @louis.
Changed5 entries
  • Paid research turns can now run up to six long-running capability calls instead of one (#940) — @louis.
  • Opening a Brain overview is now instant, even for Brains with many large documents (#919) — @opencompany-bot.
  • Claude Sonnet chats now reuse prompt context more efficiently through automatic prompt caching (#899) — @louis.
  • opencompany analytics now use a minimal PostHog setup with four allowlisted events and no autocapture, session recording, heatmaps, or message content (#923, #931) — @opencompany-bot and @louis.
  • Chat headers use tighter spacing so the conversation has more room (#936) — @opencompany-bot.
Fixed5 entries
  • Approving or declining a connected-tool action no longer repeats a successful write in the same turn (#917) — @louis.
  • Cloud Codex work now survives runner deployments, paused-sandbox resumes, and transient sandbox placement failures more reliably (#918, #921, #924, #926) — @louis.
  • Workspace invitations now open the invited workspace directly and keep the active workspace in sync after sign-in (#902) — @opencompany-bot.
  • Claude Code setup accepts tokens copied with wrapping whitespace, and its home cards now show the correct engine and status (#929, #932) — @louis.
  • Long Claude Code turns can survive repeated runner handoffs without stranding the work (#939) — @louis.
1.2.02026-07-27
Added11 entries
  • Share any opencompany chat as a live, read-only link, then stop sharing whenever you want; sharing it again creates a fresh URL so old viewers stay locked out (#892, #896) — @opencompany-bot.
  • opencompany Quick brings chat to the macOS menu bar with an Option-Command-Space shortcut (#876, #877) — @louis.
  • opencompany can browse full websites in a persistent sandbox and read up to four specific URLs per turn alongside multi-search web research (#866, #868, #881, #906) — @louis and @opencompany-bot.
  • Connected tools can now take guarded actions from chat with per-connection On, Ask, and Off controls: create Calendar events; create, update, and comment on Linear issues; send Gmail; read and edit Google Docs; read Slack; and read, update, or configure Attio lists (#842, #843, #851, #859, #870, #873, #891, #894, #895, #897, #905) — @louis and @opencompany-bot.
  • New managed research capabilities cover X, LinkedIn, YouTube, Instagram, TikTok, Semrush SEO, and structured prospect search, with a price quote and approval before paid calls (#869, #874, #887, #888) — @louis and @opencompany-bot.
  • Connect a workspace Stripe account to ask read-only questions about payments, balances, subscriptions and estimated MRR, or open invoices (#890) — @opencompany-bot.
  • Cloud Codex tasks can query their active Brain, with access checked again on every call (#884) — @opencompany-bot.
  • Ask main chat to survey Slack, Gmail, and Linear and save focused findings to the Brain with their original source attached (#838) — @louis.
  • Attach .srt subtitle files in chat, save them to the Brain, or pass the originals into Cloud Codex (#907) — @opencompany-bot.
  • New workspace owners now receive a short founder onboarding email series with one-click unsubscribe (#900) — @louis.
  • A new shareable use-cases page shows what opencompany can do (#885) — @louis.
Changed5 entries
  • The Slack answer bot now runs through the full opencompany agent, understands follow-up threads, can work across multiple Brains, and shows live status while it works (#854) — @louis.
  • You can keep drafting your next message while opencompany responds, tool rows are less cluttered, and the context-window meter updates throughout multi-step turns (#857, #871, #872, #904) — @opencompany-bot.
  • Onboarding now guides connected sources all the way to actively feeding the Brain, and valid company URLs are no longer blocked by temporary reachability checks (#852, #883) — @louis and @opencompany-bot.
  • Brain queries now default to curated pages, paginate consistently across chat, MCP, and the CLI, and keep raw evidence available when you explicitly need it (#880) — @louis.
  • Basic Brain ingestion is more efficient, groups each GitHub pull request lifecycle into one filing pass, and preserves valid source citations (#846, #847, #848, #849, #850, #853) — @louis and @opencompany-bot.
Fixed4 entries
  • Accepted workspace invitations now appear after sign-in even when you already belong to another workspace (#889) — @opencompany-bot.
  • Brain sidebar context-menu actions work with the mouse again, and each chat keeps its original model instead of inheriting a choice from another tab (#864, #865) — @louis and @opencompany-bot.
  • Cloud Codex handoffs and long, parallel, or paid chat actions recover more reliably instead of timing out or rejecting valid tool calls (#858, #862, #875, #878, #879, #882, #886) — @louis.
  • Brain tool results no longer appear twice in the model's chat context (#903) — @louis.
1.1.02026-07-22
Added12 entries
  • Four new brain sources file in automatically: Attio, HubSpot, Fathom, and Granola — so your CRM records, meeting notes, and calls land in the brain without you lifting a finger (#776, #757, #767, #758) — @louis.
  • Ask opencompany right from Slack — mention @opencompany in a channel and it answers from your brain (beta) (#759) — @louis.
  • opencompany chat can now search across your connected tools — Slack, Gmail, GitHub, Linear, Attio, Google Drive, and Google Calendar — and pull the results straight into the conversation (#830, #828, #832, #831, #807) — @louis.
  • Chat answers now cite the brain sources they came from, so you can see where an answer is grounded (#775) — @louis.
  • Codex coding sessions run in the cloud and show up as Tasks, with model selection and file attachments (#784, #766, #771) — @louis.
  • The changelog now lives in the sidebar, so you can see what's new without leaving the app (#833) — @louis.
  • Browse and restore your archived chats from the ⌘K palette (#796) — @louis.
  • opencompany remembers the model you picked for main chat between sessions (#829) — @louis.
  • Kimi K3 is now in the model picker (#777) — @louis.
  • A sidebar feedback widget sends your bugs and ideas straight to our team (#764) — @louis.
  • Save anything to your brain over MCP from any agent (#748) — @louis.
  • Migrated the opencompany blog archive into the new marketing site, keeping its existing URLs and search metadata (#765) — @louis.
Changed4 entries
  • Billing is now a pure pay-as-you-go wallet — you pay only for what you use and top up yourself, with no seats or plans (#802) — @louis.
  • The integrations settings page was redesigned as a clean grid of brand cards, with a scope switcher and the option to connect tools during onboarding (#810, #824, #834) — @louis.
  • Brain ingestion is more reliable, resumes cleanly after upgrades, and brain search now returns sharper, better-ranked results (#792, #770, #835) — @louis.
  • A context-usage tooltip shows how full the model's context window is, and switching between chats is now instant (#795, #804) — @louis.
Fixed5 entries
  • Linear now reconnects cleanly instead of getting stuck showing as disconnected (#814) — @louis.
  • Codex chat sessions no longer get stuck when queued, closed, or interrupted by a deploy (#794, #798, #812) — @louis.
  • Attio system noise is kept out of the brain, and Attio ingestion is more robust (#819, #781) — @louis.
  • Integration setup failures now show a clear error instead of failing silently (#821) — @louis.
  • Checkout is more reliable, with fixed promotion-code and idempotency handling (#761, #763) — @louis.
1.0.02026-07-15
Added12 entries
  • opencompany is here: a company brain that files what matters from your team's tools and conversations, links it to what you already know, and serves it to any agent over MCP (#554) — @louis.
  • opencompany Brain is a typed, folder-organized knowledge base with an evidence taxonomy, wiki-style links rendered as chips, and first-class binary documents including PDFs (#558, #607, #611, #615, #620, #613) — @louis.
  • Brains ingest from your sources automatically — Jamie meeting notes, GitHub (PRs, issues, comments), Gmail, Linear, Google Drive, X, and chat captures — filing and linking what matters (#575, #652, #659, #641, #715, #573, #631) — @louis.
  • Ingestion can enrich entries with web search, respects per-brain spend budgets, and imports company context during onboarding (#691, #725, #724, #728) — @louis.
  • opencompany chat streams live with disconnect-safe, resumable sessions, file attachments that flow end-to-end into the brain, chat pinning, background chats, and Cmd+K to start a new one (#633, #700, #731, #577, #627) — @louis.
  • Codex-backed sessions run in opencompany across cloud and local sandboxes, with in-chat controls, sandbox status, and per-task engine selection (#555, #649, #624, #662, #661) — @louis.
  • Tasks can run on a schedule, notify their originating chat on completion, expose their harness config, and sit behind a background-task preference (#560, #557, #559, #717) — @louis.
  • A single user-level MCP endpoint at /mcp exposes every brain you can access, with a personal MCP settings tab and member onboarding (#739, #600, #714) — @louis.
  • Workspaces gate brain access with member-level permissions, member-contributed sources, multi-account connections, and cross-member deduplication (#596, #656, #740) — @louis.
  • Workspace billing ships with pooled monthly allowances, per-seat pricing, USD credits, per-brain intelligence tiers, credit top-up promotion codes, and a spend overview with charts (#718, #732, #742, #745, #648, #651) — @louis.
  • A brain overview dashboard, activity feed, and settings overview give each brain a home, alongside a model picker, Fumadocs-powered docs, browser research tools, an onboarding flow, and new app icons (#741, #625, #669, #594, #572, #576, #710, #727) — @louis.
  • Codex chats get auto-generated titles, and opencompany gained an appearance theme toggle and a quicker new-brain action in the sidebar (#668, #703, #702) — @opencompany-bot.
Changed4 entries
  • Brain retrieval was rebuilt on a Postgres-native read plane and unified into a single surface; chat and MCP now read the brain read-only, with all writes flowing through ingestion (#621, #695, #690) — @louis.
  • Navigation across the brain and app is now client-first for instant transitions (#591, #635) — @louis.
  • The sidebar gained a workspace switcher in its header, recent-chat history, archive-on-hover, and a per-brain context menu (#667, #696, #698, #712) — @louis.
  • The ingestion model was updated, the per-event ingestion budget was raised to one dollar, and the browser tab now reads "opencompany" (#655, #734, #747) — @louis.
Fixed4 entries
  • Brain ingestion no longer suffers retry storms or conflict-page pollution, and retrieval tracking is accurate (#707, #743) — @louis.
  • Fixed an infinite render loop that froze brain navigation and a composer freeze when input scrolled past its max height (#736, #730) — @louis.
  • opencompany chat session continuity and post-completion navigation are reliable, and chat pin transactions are consistent (#590, #706, #737) — @louis.
  • Jamie webhook API-key handling and setup were corrected (#580, #654) — @louis.
0.18.02026-06-25
Added9 entries
  • Agents can now run Codex-backed sessions with persisted auth, sandbox lifecycle support, runtime event ingestion, and transcript UI support behind the Codex engine flag (#525) — @louis.
  • Google Drive is now available as a first-party integration with OAuth, Drive search/read/export tools, and Google Docs create/update tools (#515) — @louis.
  • Company workspaces now have a dedicated Integrations tab for first-party and MCP integration controls (#522) — @louis.
  • Personal routines now include scheduling with create, edit, delete, toggle, run-now, next-run, and timezone controls (#458, #507) — @louis.
  • Billing now supports weekly spending limits and automatic credit refills with saved cards (#505) — @jasper.
  • Delegated coding and memory tools now use a runner-hosted LLM token broker, keeping raw provider keys out of sandboxes while metering brokered usage server-side (#439) — @louis.
  • Z.ai GLM-5.2 is now available in the model catalog with billing and reasoning-capable model support (#486) — @opencompany-bot.
  • Local development can now persist Turbo logs and inspect them through bun run dev:logs (#518) — @louis.
  • Codex E2B toolbox templates now include Playwright and Playwright-managed Chromium for browser-ready coding sessions (#525) — @louis.
Changed6 entries
  • Mobile Memory pages now let users swipe the file tree away for a full-width file view, while guarding iOS edge-back gestures (#510) — @jasper.
  • Tool-call cards now show the brand icon for external services like Linear, Slack, Gmail, GitHub, Notion, PostHog, Google Calendar, Google Drive, and Neon (#524) — @jasper.
  • After-session memory updates now appear as a compact footer badge with a tooltip and link to the spawned memory-pass session (#526) — @louis.
  • The personal and company sidebar space switcher is now a dropdown, and the private beta badge has been removed (#520) — @louis.
  • Feedback reports now use a one-tap Bug, Feedback, and Idea segmented selector with Feedback selected by default (#499) — @jasper.
  • Company and personal route navigation now uses faster links, debounced prefetching, and loading skeletons for slow routes (#523) — @louis.
Fixed9 entries
  • Google Drive OAuth now uses the narrower drive.file scope so production connections are not blocked by full-Drive restricted-scope verification (#517) — @louis.
  • iOS PWA layouts now respect safe-area insets, avoid keyboard popups in the mobile model picker, use full-width chat messages on phones, and prevent mobile settings clipping (#501, #504) — @jasper.
  • MCP reconnect and tool-discovery failures now return recoverable tool outputs instead of failing the whole session (#503) — @louis.
  • Durable Stream publishing now retries with a fresh producer when a cached producer has already closed, improving live delivery for lifecycle and memory events (#502) — @louis.
  • The automatic-refill Add a card flow now passes the required Stripe Checkout currency (#511) — @jasper.
  • Personal session unread dots are visible again and are no longer re-armed by background after-session memory passes (#509, #513) — @jasper.
  • Existing personal agents now backfill missing soul.md files so the Soul nav opens the editor instead of jumping to a new chat (#514) — @jasper.
  • Company sidebars no longer show personal-agent threads in recent or pinned session history (#521) — @louis.
  • Reading position is preserved when completed work blocks collapse or expand in long session transcripts (#508) — @jasper.
0.17.02026-06-22
Fixed1 entry
  • Completed assistant replies now appear live when the final stream payload is missed, without needing a page reload (#496) — @louis.
0.16.02026-06-19
Added2 entries
  • Personal sessions on mobile can now swipe from the opposite edge for session details, while filmstrip gestures respect swipe direction (#492, #495) — @jasper.
  • The personal agent sidebar now includes a tabbed session details inspector (#489) — @louis.
Changed4 entries
  • Feedback reports now submit optimistically and archive actions show a confirmation toast (#491) — @jasper.
  • Personal Brain and after-session memory guidance is stricter, so agents only save persistent files or memories when user intent is clear (#485) — @louis.
  • The personal sidebar home entry is now labeled New Session and uses the send icon (#487) — @louis.
  • Personal sidebar session lists now hide scrollbars while remaining scrollable (#494) — @jasper.
Fixed2 entries
  • Agent replies no longer reorder in the transcript when stream events arrive out of order (#493) — @jasper.
  • Mobile drawer swipes now work more reliably in Safari while preserving normal vertical page scrolling (#490) — @jasper.
0.15.02026-06-18
Added2 entries
  • Personal Brain now keeps restorable file versions, with restore tooling, caps, and runner guidance for safer durable writes (#483) — @jasper.
  • Mobile users can now swipe to open the main navigation drawer (#484) — @jasper.
Fixed4 entries
  • The onboarding wizard no longer resets when setting tool policies during the capabilities step (#481) — @jasper.
  • The three-dot menu in the brain file view no longer renders behind the file text (#480) — @jasper.
  • GitHub work repositories granted after connecting now resolve via a live lookup instead of being missed (#479) — @jasper.
  • Various mobile and phone UI fixes (#478) — @jasper.
0.14.02026-06-16
Added5 entries
  • OpenRouter Fusion models are now supported (#473) — @louis.
  • Messages sent mid-run now offer Steer, Queue, and Interrupt send-modes (#468) — @jasper.
  • Chat sessions can now be renamed via double-click or a right-click menu (#467) — @jasper.
  • Clarifying questions now always include a free-text "Other" answer (#466) — @jasper.
  • Screenshots dropped into chat are now attached to the Linear issue created from that conversation (#464) — @jasper.
Changed3 entries
  • Built-in skills now show distinct icons instead of a shared Sparkles placeholder (#475) — @jasper.
  • Memory lookup prompting tightened for more relevant recall (#474) — @louis.
  • Typed @mentions in the agent editor are no longer auto-converted as you type (#470) — @jasper.
Fixed4 entries
  • Near-identical person records in memory are now deduplicated (#477) — @jasper.
  • The thinking trace now separates distinct reasoning phases instead of running them together (#476) — @jasper.
  • The "Updated memory" card no longer overlaps the message footer (#469) — @jasper.
  • Fixed the personal billing redirect (#465) — @louis.
0.13.02026-06-12
Added8 entries
  • Agents can now create Linear issues directly from chat, preserving the relevant conversation context in the issue (#462) — @jasper.
  • Notion is now available as a first-party MCP integration with OAuth connection flow and tool discovery (#455) — @louis.
  • New users can now complete a Leo-led onboarding flow with updated personal onboarding screens and kickoff behavior (#451) — @louis.
  • Agents can now use the bundled YC knowledge skill (#456) — @louis.
  • Feedback reports can now include screenshots, which are uploaded and embedded inline in the linked Linear issue (#447) — @jasper.
  • The personal Memory surface is now URL-addressable, so selected memory paths can be opened and restored from direct links (#446) — @jasper.
  • Memory queries now support relative --since windows and query-less recent-memory listings (#444) — @louis.
  • Preview workspaces now show an early-preview badge in the sidebar (#457) — @louis.
Changed3 entries
  • Personal sidebar navigation was refreshed for the updated personal surface structure (#449) — @louis.
  • GitHub permission scopes were consolidated and renamed around clearer permission groups (#450) — @louis.
  • Signup welcome email delivery and unsubscribe handling were refactored for more reliable account email flows (#452) — @louis.
Fixed4 entries
  • Home-space image drag and drop now preserves sent images and shows them in the chat bubble after sending (#454) — @jasper.
  • Sessions now recover stale stream state after returning to a tab instead of staying stuck on outdated loading state (#448) — @louis.
  • Personal Brain writes are no longer silently dropped when the agent bundle exceeds the 80-file cap (#445) — @louis.
  • Chat scrolling now settles at the bottom when expected during active sessions (#431) — @jasper.
0.12.02026-06-11
Added10 entries
  • Brain files created or edited by agents now show as clickable attachments under assistant turns, and the company Brain can be opened directly at /brain/<path> deep links (#356) — @louis.
  • The personal Brain is now URL-addressable too, with nested file links restoring the selected file on reload and keeping the Brain navigation active (#426) — @jasper.
  • Personal Brain files created or edited during agent turns now link directly from the session transcript into the matching personal Brain path (#442) — @louis.
  • Personal sessions now support split-screen panes, so multiple chats can stay visible and be arranged side by side (#441) — @louis.
  • Agents can now start scoped subagents with selected tool grants, with subagent lifecycle events shown in the parent session transcript (#421) — @louis.
  • The personal Integrations tab now expands inline with connected account details, permission summaries, degraded-access warnings, disconnect actions, and GitHub repository refresh (#390) — @louis.
  • Personal recall can now be constrained by time windows, including query-less recent-session snippets, so agents can ask for memory from a specific lookback period (#416) — @louis.
  • Large pasted text is captured through the attachment pipeline, and oversized text attachments are materialized as sandbox file references instead of being inlined into every model call (#410) — @louis.
  • Session usage now updates live during streaming, long-running tool calls show an elapsed counter, and the context-window tooltip includes session cost (#395, #376, #399) — @louis.
  • Runner observability now records per-tool-call phase timings, Braintrust spans, PostHog rollups, and hidden sandbox hydration timing events for production latency diagnosis (#411, #417) — @louis.
Changed10 entries
  • Starting a session from the personal home prompt now switches to an optimistic session view immediately while creation and navigation finish in the background (#420) — @louis.
  • Runner sandbox hydration now overlaps bundle, skill, and setup work to reduce startup latency for sessions that need a sandbox (#434) — @louis.
  • Personal inbox cards now render markdown in their body and step text, matching the markdown contract used by the inbox tool (#438) — @louis.
  • Recall recap guidance is tighter, with cleaner formatting for recalled memories and follow-up summaries (#436) — @louis.
  • Preview deployment PR comments now include ?skipOnboarding, letting reviewers test fresh preview accounts without going through onboarding (#435) — @louis.
  • New personal agents now default to Leo on Kimi K2.6, with updated default capabilities, model selection, settings reset, and personal agent source normalization (#409, #425) — @louis.
  • Memory tooling now returns richer memory query and memory get output, clearer invalid-usage help, a tighter keeper kickoff prompt, and canonical guidance that durable facts belong in structured memory while agent/user.md is the always-loaded digest (#413, #418, #422, #428) — @louis.
  • Memory keeper passes now use Gemini 3.1 Flash Lite, surface what they stored, and appear in the transcript near the turn they followed instead of pinned to the bottom (#401, #397) — @louis.
  • Runtime details now stay collapsed by default for new sessions, including sessions with related parent or child metadata (#423) — @louis.
  • Slack support channel names now keep the per-workspace suffix at the end, for example <customer>-x-opencompany-<id8> (#394) — @jasper.
Fixed8 entries
  • Slash command session starts now route to the intended personal or company surface instead of opening the wrong session view (#432) — @louis.
  • Session lists now consistently sort by creation time across the company and personal sidebars (#437) — @louis.
  • Personal agent skill slash commands are available again in the composer slash menu and insert the right skill mention (#440) — @louis.
  • Personal-session inspector links now stay inside the personal surface instead of opening related sessions, session pages, or agent links under /company (#391) — @louis.
  • Personal runtime tool guidance now advertises the correct work/, personal-brain/, and agent/ roots and blocks generic file tools from editing structured memory paths (#414) — @louis.
  • OpenCode tool usage is now parsed from current step_finish events and charged back to the spawning session instead of silently appearing as zero-cost work (#402) — @jasper.
  • Runner shutdown and interruption handling now drains in-flight runs, persists abort/failure state more reliably, and avoids replaying interrupted deploy-time work (#378) — @louis.
  • E2B stream callbacks are guarded against detached rejections so stopping an opencode_coder command no longer crashes the runner process (#400) — @louis.
0.11.02026-06-10
Added6 entries
  • The new /personal experiment surface gives each user a private default agent with its own sessions, behavior editor, capability panels, context files, memory, tools, skills, integrations, and settings views (#353) — @louis.
  • Claude Fable 5 is now available end to end in the model catalog, with reasoning, image, PDF, long-context, and billing support (#384) — @louis.
  • Preview deployments can now complete Gmail and Google Calendar OAuth through a stable callback broker, making Google integration testing work across hosted preview URLs (#382) — @louis.
  • Public changelog entries can now include embedded images and short looping videos, with docs for recording, compressing, uploading, and referencing changelog media (#387) — @louis.
  • The account menu now shows when the currently running build was last deployed, with release details in the tooltip (#388) — @louis.
  • Personal agents now support plain @github for all installation repositories and repo-scoped @github/owner/repo mentions, with matching sandbox and coding-tool access (#392) — @louis.
Changed1 entry
  • Sandbox preparation no longer waits on optional developer-tool installation during the critical setup path (#393) — @louis.
Fixed1 entry
  • Personal recall searches no longer fail before running when the runner configures its statement timeout (#389) — @louis.
0.10.02026-06-08
Added3 entries
  • Paste, drag, drop, or pick screenshots, PDFs, and text/code files in the session composer; attachments are validated per model, stored in private Vercel Blob, rendered as compact cards, and inlined into runner model calls (#362) — @jasper.
  • Attached skills now appear in the composer slash menu as /<command> entries, using the skill's declared command frontmatter when present and inserting the matching @skill/<id> mention when selected (#350) — @louis.
  • Brain pages now show when the file was last updated directly in the header, and include a copy button for the page title plus current contents (#355, #358) — @louis, @jasper.
Changed2 entries
  • Session pages now use the session title as the browser tab title, making multiple open chats easier to tell apart (#354) — @louis.
  • Slack Connect support channels now use safer <customer>-<id8>-x-opencompany naming, include recovery sweeps for stuck provisioning, and have expanded setup and operations docs (#335) — @jasper.
Fixed3 entries
  • opencode_coder timeouts and command failures now preserve partial diffs, persist artifacts, return a resumable opencode session id, and report timeout status instead of dropping the work (#344) — @louis.
  • Long-running tool calls no longer trigger runner job replay and double billing after transient lease gaps; job lease TTL and lease-busy retry ceilings are now configurable (#345) — @louis.
  • Coding sandboxes now configure GitHub git credentials and best-effort ensure rg and bun are available, so plain git commands, repo search, and tests work more reliably inside sessions (#346) — @louis.
0.9.02026-06-05
Added7 entries
  • Gmail and Google Calendar are now first-party integrations — connect one or more Google accounts to give agents read-only Gmail and read/write Calendar tools, with per-account calendar selection (#319, #320) — @louis.
  • Switch the model for a single chat on the fly without changing the agent's saved default, in a redesigned composer with a toolbar tray for the agent and model selectors (#331) — @louis.
  • A context-window usage ring in the session top bar shows how full the model's context window is for the current turn (#332) — @louis.
  • Slack Connect — completing onboarding now provisions a private shared Slack channel with our team, surfaced as a "Connect on Slack" card on your workspace home (#278) — @jasper.
  • Tool calls that touch a Linear issue now show an external-link icon that opens the issue directly (#324) — @jasper.
  • Deferred tool discovery — capability tools and coding agents are no longer all preloaded; the model discovers them on demand via find_tools and runs them through a use_tool dispatcher, improving tool selection and shrinking the cached prompt (#321) — @louis.
  • New shared @opencompany/ui component package (shadcn / Tailwind v4) and a design-system app to document and showcase it (#323) — @louis.
Changed7 entries
  • Deferred-tool arguments now run through a validate → coerce → repair pipeline before dispatch, so smaller models' malformed tool calls are fixed automatically instead of failing (#328) — @louis.
  • Brain access scope is now enforced at the tool layer, so out-of-scope Brain writes are rejected up front with a clear error instead of being silently dropped at sync (#327) — @louis.
  • The new-session empty state no longer shows default prompt chips (#315) — @louis.
  • The /btw start toast no longer echoes the prompt; it shows just a confirmation and an Open action (#330) — @louis.
  • Faster navigation — removed the route-level session loader that flashed skeletons on fast nav (#303) — @louis.
  • Production deploys are faster and now recover gracefully from interrupted runner deploys (#310, #318) — @louis.
  • Runner model and tool calls are now traced via Braintrust's wrapAISDK defaults for richer, lower-maintenance observability (#304) — @louis.
Fixed3 entries
  • Unknown-tool and invalid tool-input calls are now recovered as tool results that steer the model back on track, instead of crashing the turn (#314, #325) — @louis.
  • Fixed a startup-status race that briefly showed "Stopped before finishing" on a freshly started session that was actually still running (#306, #312) — @louis.
  • Runner completions that contain only reasoning and no answer are now rejected instead of recorded as a successful turn (#300) — @louis.
Security1 entry
  • Hardened GitHub permission gating for agent shell commands, closing permission bypasses (#302) — @louis.
0.8.02026-06-04
Added8 entries
  • Real-time session sync and durable token streaming powered by Electric and Durable Streams, so chat updates propagate live and survive interrupted connections (#283) — @louis.
  • New opencode coding agent tool, letting agents run coding tasks through opencode (#266) — @louis.
  • Official PostHog MCP support with OAuth / Dynamic Client Registration, so agents can connect to PostHog without manual token wiring (#269) — @louis.
  • E2B sandbox compute is now metered and billed as agent usage (#264) — @louis.
  • Upload a custom profile picture for your account (#257) — @jasper.
  • Credit top-up analytics to track balance refills (#279) — @louis.
  • opencompany icon assets (#284) — @louis.
  • External skills support (V1) — attach a public GitHub repo or skills.sh page to an agent via an @skill mention; the skill is snapshotted workspace-side and materialized read-only into the runner sandbox at session start alongside built-in skills (#296) — @louis.
Changed9 entries
  • Session detail now paints instantly from local TanStack DB collections instead of waiting on the server (#290) — @louis.
  • Improved AI SDK prompt caching for lower latency and cost on repeated context (#275, #285) — @louis.
  • The agent self-edit skill now asks before making changes, stays lightweight, and better equips coding agents (#281) — @louis.
  • Onboarding call step reworked so its actions fit on a single screen (#263) — @jasper.
  • Web functions now run in the Frankfurt region for lower latency to European users (#286) — @louis.
  • MCP server tools are now lazy-loaded via per-server search_tools / use_tool meta-tools, so raw tool schemas are only sent to the model on demand instead of being injected on every turn — eliminating the per-turn token cost of large MCP catalogs like Slack (#298) — @louis.
  • Starting a new session from the root-route prompt now navigates instantly; session detail is assembled in-memory from the just-inserted rows instead of waiting on seven follow-up database reads (#295) — @louis.
  • E2B sandbox warm-up now begins at the start of a turn so the ~10 s cold-start overlaps the model's initial token stream rather than blocking the first tool call (#297) — @louis.
  • Runner parallel-session concurrency is now configurable via RUNNER_WORKER_CONCURRENCY (default 8); production limit raised to 40 (#294) — @louis.
Fixed10 entries
  • Coalesced durable-stream token appends to fix streaming lag in production (#288) — @louis.
  • Electric and Durable Streams now work in local development (#291), with producer linger and in-flight limits tuned for throughput (#292) — @louis.
  • Tool output previews are now redacted of secrets before being persisted to the database (#273) — @louis.
  • Approval stream state no longer gets stuck out of sync (#272) — @louis.
  • opencode now resolves the correct attached repository (#271) — @louis.
  • Linear save_* MCP tools are no longer misclassified as Admin-only (#267) — @louis.
  • @mention pills no longer break after an agent edits itself (#268) — @louis.
  • A run is never replayed after its lease expires, and onboarding dedup is hardened against duplicate runs (#262) — @louis.
  • The Render deploy trigger is now resilient to empty API responses (#287) — @louis.
  • Session status and error no longer flicker blank at the start of a session; the live stream only takes authority over the Postgres snapshot once it has received a status-bearing event (#299) — @louis.
0.7.02026-06-03
Added5 entries
  • Agents can now pause mid-run to ask structured questions with single-select, multi-select, and "Other" answers, then continue from the user's response (#255) — @louis.
  • Onboarding agents now use structured questions and receive signup names in their starting context, making first-run setup more personal and less brittle (#259, #260) — @louis.
  • Apify-backed Instagram and TikTok tools for public profiles, posts, reels, videos, comments, search, and async social scraping jobs (#253) — @louis.
  • Searchable, provider-grouped model picker with Capability, Speed, and Cost ratings so models are easier to compare at a glance (#254) — @louis.
  • Day-one Brain wiki and private agent agent/soul.md scaffolding, giving new workspaces a cleaner shared knowledge structure and better self-configuration defaults (#251) — @louis.
Changed2 entries
  • Onboarding call booking now uses one booking-aware button, the official Cal.com embed bootstrap, and clearer new-tab labeling (#249, #250, #258) — @jasper.
  • Public changelog entries now include author attributions (#256) — @louis.
Fixed4 entries
  • Chat scroll-to-top behavior after sending a message is now stable during streaming and viewport resizing, with less jitter and less fighting user scrolls (#248) — @jasper.
  • Paused sessions now refresh stream tokens correctly, and settings hydrate consistently after reloads (#252) — @louis.
  • Editing a markdown title or header no longer flickers while typing (#201) — @jasper.
  • Naming a new Brain file from its tab now updates the sidebar tree immediately (#198) — @jasper.
0.6.02026-06-02
Added11 entries
  • Guided first-run setup — after signing up you land in a live setup conversation that helps configure your workspace, instead of an empty editor (#241) — @louis.
  • New /btw command to start a new session without leaving your current view (#243) — @louis.
  • New /clear command in the chat composer (#231) — @louis.
  • "Run now" button to trigger a scheduled agent immediately (#236) — @louis.
  • Agents can now pull content from TikTok and Instagram (#239) and YouTube (#233) — @louis.
  • Pin sessions to the top of the sidebar (#213) — @jasper.
  • Live product status indicator in the sidebar, linking to the public status page (#224) — @louis.
  • Agents can now adjust their own schedules (#217) — @louis.
  • Workspace tool permissions — allow, ask, or deny which tools and integrations an agent can use, with approval prompts that are remembered between runs (#229) — @louis.
  • Live reasoning support for Kimi models (#226) — @louis.
  • Provider logos for xAI and MiniMax in the model picker (#232) — @jasper.
Changed5 entries
  • Agent runs no longer break when an integration like Linear or Slack is turned on but not yet connected — the agent now asks you to connect it instead (#238) — @louis.
  • When you send a message, it scrolls to the top of the chat, matching the ChatGPT/Claude experience (#219) — @jasper.
  • The copy button now stays visible on assistant replies and appears on hover for your own messages (#222) — @jasper.
  • The onboarding call step now has a clear "Skip for now" button (#220) — @jasper.
  • The "reconnecting" banner no longer appears in chat during normal streaming; connection status now lives in the inspector instead (#242) — @louis.
Fixed6 entries
  • Linear no longer shows as disconnected after you reconnect with valid credentials (#234) — @louis.
  • Agent turns that quietly stop mid-task are now flagged instead of being recorded as a silent success — important for scheduled and unattended runs (#225) — @louis.
  • Sessions now recover when a response stream gets interrupted (#223) — @louis.
  • Rapidly toggling a session's pin no longer causes glitches (#218) — @jasper.
  • Restored the after-session hook menu when typing at the start of a line (#227) — @jasper.
  • The hook suggestion popup now stays beneath the model dropdown (#230) — @jasper.
0.5.02026-06-01
Added7 entries
  • Cron-based agent schedules — agents can now be scheduled to run on a recurring basis (#208) — @louis.
  • Read-only X (Twitter) hosted tool for agents (#212) — @louis.
  • Agent self-update skill, allowing agents to evolve their own .agent definition (#210) — @louis.
  • Dark mode support (#186) — @louis.
  • Grok model support via the AI gateway (#206) — @louis.
  • MiniMax and Kimi gateway models (#193) — @louis.
  • Client-side felt time-to-first-token measurement (#192) — @louis.
Changed5 entries
  • GitHub repositories are now decoupled from the amp tool, making repository bindings more flexible (#204) — @louis.
  • Lazy GitHub sandbox initialization is more solid and reliable (#207) — @louis.
  • Vercel Skew Protection is enabled via custom deploymentId for safer deploys (#205) — @louis.
  • Onboarding call booking step is now skippable (#191) — @jasper.
  • Assistant turn duration is now shown next to the copy button (#187) — @jasper.
Fixed6 entries
  • Agents no longer disappear from the list after creating or editing a new agent (#203) — @jasper.
  • Renaming an agent no longer resets it to "Untitled agent" (#194) — @jasper.
  • Vercel deployment ID length is correctly handled (#211) — @louis.
  • Session archiving is now optimistic and instant instead of waiting on the server (#200) — @jasper.
  • Feedback dialog auto-closes after a successful submit (#190) — @jasper.
  • Page no longer rubber-bands at the fold due to overscroll-behavior fix (#189) — @jasper.
0.4.12026-05-30
Changed3 entries
  • Runner job worker now wakes immediately on enqueue instead of waiting for the next poll interval, reducing time-to-first-token for agent sessions — @louis.
  • Web dispatch no longer blocks on analytics flush; PostHog capture and runner dispatch now run concurrently — @louis.
  • Workspace context loading collapses user, workspace, and role into a single joined query, and session submission runs the balance check and auth lookup concurrently, cutting pre-dispatch database round-trips — @louis.
Fixed1 entry
  • Runner event serialization no longer crashes when created_at arrives as a string from the raw lease-write path; the value is now coerced to a Date at the source — @louis.
0.4.02026-05-28
Added7 entries
  • Workspace integration resource bindings with encrypted credential storage, multi-connection GitHub repository binding, and runtime validation for bound resources.
  • Linear and Slack MCP support, including workspace setup flows, agent editor tool gating, runner-side MCP execution, and .agent file configuration.
  • Delegated agent sessions through @agent/<slug> mentions and a delegate_to_agent tool, with child-session tracking and parent usage rollups.
  • Repo-scoped GitHub auth for agent shell commands and AMP runs, with ephemeral credentials and output redaction.
  • GPT 5.2 Codex support and configurable AMP execution modes.
  • Agent editor support for markdown headings, ordered and unordered lists, and automatic conversion of typed or pasted @/# mentions into pills.
  • A richer chat composer with quick-start chips, auto-resize, stop controls, copy-message actions, drag-and-drop and paste handling shells, and improved keyboard hints.
Changed3 entries
  • Local setup now keeps WorkOS authentication redirects on localhost:3000 while continuing to use ngrok for GitHub and other public integration callbacks.
  • Production web deploys now rely on the GitHub Actions release workflow instead of Vercel Git-triggered deployments.
  • AMP command handling now avoids rejected stream-JSON mode combinations and falls back to sanitized plain output when structured results are unavailable.
Fixed5 entries
  • GitHub integration sync in production now avoids unsupported Neon HTTP transactions and normalizes legacy agent configs.
  • Sandbox repository clones, AMP publishing, and authenticated shell commands now handle GitHub App credentials more reliably.
  • Session chat no longer allows duplicate submissions while an assistant is responding and no longer leaves failed sessions stuck in a thinking state.
  • The New agent button now shows immediate pending feedback, blocks double-click creation, and surfaces creation failures.
  • Linear MCP credential setup now reports missing encryption-key configuration with clearer diagnostics.
0.3.02026-05-27
Added6 entries
  • Deterministic edit_file support for hosted agents, giving runner sessions a safer and more reviewable way to modify files.
  • Expanded hosted Exa tools and clearer search schema guidance for agent research workflows.
  • Common Vercel AI Gateway model presets across the editor, runtime configuration, and billing calculations.
  • Model and turn-cost analytics for better visibility into session usage and credit spend.
  • Signup welcome emails, session-aware feedback reports, and an onboarding call booking step.
  • A pulsing sidebar indicator for active sessions.
Changed4 entries
  • Runner internals now separate session lifecycle, job leasing, tool dispatch, model streaming, and usage recording for more reliable hosted runs.
  • Agent file parsing, mention handling, runtime types, and after-session behavior now live in the shared agent runtime package used by both web and runner.
  • GitHub sync jobs now have retry and sweeper support to recover pending workspace file updates more reliably.
  • Development and review tooling now includes CodeRabbit configuration and more consistent local script environment loading.
Fixed4 entries
  • Blank user messages now survive session payload handling and runtime event rendering.
  • Agent detail mention caches no longer leak across agent boundaries.
  • Hosted tool fan-out is capped to avoid runaway parallel tool execution.
  • GitHub integration callback diagnostics now expose enough detail to troubleshoot failed setup flows.
0.2.02026-05-26
Added7 entries
  • Git-backed agent editing, including file-backed .agent sync, rename propagation, and visible GitHub sync state.
  • Hosted agent sessions with root-route session start, run controls, archive lifecycle, automatic sandbox pausing, and first-message session titles.
  • Agent tools for hosted Exa search and lightweight web fetching.
  • WorkOS organization workspace support, signup onboarding, default workspace credits, and personal environment setup.
  • Brain context files for agents, plus a more complete workspace file experience.
  • Session usage, billing, analytics, and cached-token tracking so teams can understand credit spend.
  • Public docs, public changelog, and in-app feedback intake.
Changed5 entries
  • Workspace sessions now load from cache first and route transitions respond faster.
  • Session UI now better explains reasoning model choice, token usage, tool replay, and follow-up state.
  • Workspace loading and agent navigation states are more predictable.
  • Release automation now coalesces CI, syncs Inngest production functions, records GitHub deployments, and runs on Blacksmith.
  • Observability now covers agent sync, launch events, runner logs, and Better Stack error context.
Fixed5 entries
  • AuthKit sign-in and sign-up redirects now use full document navigation when required.
  • Runner sandboxes recover more reliably from E2B lifecycle and tool failures.
  • Brain new file and new folder actions no longer throw on click.
  • Direct session title generation and tool replay behave consistently.
  • Production release smoke checks and Render release waiting are more reliable.
Removed1 entry
  • Persisted tool delta events and the sidebar running badge to reduce noisy state.
0.1.02026-05-20
Added3 entries
  • Initial Next.js app prototype with sidebar navigation, Agents, Brain, Inbox, Settings, and session surfaces.
  • Early WorkOS authentication, branchable Neon database setup, and persistent agents.
  • Turborepo workspace structure with Bun, shared database package, CI, formatting, tests, and secret scanning.
Releases
1.31.02026-09-181.30.02026-09-171.29.02026-09-161.28.02026-09-151.27.02026-09-141.26.02026-09-111.25.02026-09-101.24.02026-09-071.23.02026-09-041.22.02026-09-031.21.02026-09-011.20.02026-08-281.19.02026-08-271.18.02026-08-241.17.02026-08-211.16.02026-08-201.15.02026-08-171.14.02026-08-141.13.02026-08-121.12.02026-08-111.11.02026-08-101.10.02026-08-071.9.02026-08-061.8.02026-08-051.7.02026-08-041.6.02026-08-031.5.02026-08-011.4.02026-07-291.3.02026-07-281.2.02026-07-271.1.02026-07-221.0.02026-07-150.18.02026-06-250.17.02026-06-220.16.02026-06-190.15.02026-06-180.14.02026-06-160.13.02026-06-120.12.02026-06-110.11.02026-06-100.10.02026-06-080.9.02026-06-050.8.02026-06-040.7.02026-06-030.6.02026-06-020.5.02026-06-010.4.12026-05-300.4.02026-05-280.3.02026-05-270.2.02026-05-260.1.02026-05-20