Skip to content

AI-powered virtual executive team — a single coherent executive persona backed by 8 specialist agents (FastAPI + Next.js).

111 KB
Raw Blame

Changelog

All notable changes to this project are documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

From 0.3.0 on, entries are written by release-please from the titles of the merged pull requests (feat → Added, fix → Fixed). The open release PR holds the next entry; edit it there before merging if a line needs rewording.

0.6.7 (2026-10-27)

Added

  • briefing: open and reply to Right now's email points from Home (#651) (ddc4429)
  • briefing: show your own inbox in Right now and message you when it changes (#640) (428456a)
  • delegation: open reply cards as a conversation and redraft on newer mail (#639) (bd293a7)
  • orchestrator: answer on a pushback model after the person pushes back (#653) (3eb8242)
  • ui: warn when the host's user isn't who is signed in, and let hosts read who that is (#638) (4c5e154)

Fixed

  • attunement: raise the open-loop caps (#642) (fc0ea3c)
  • briefing: drop answered mail from Right now and show its steps while it updates (#652) (297cbe0)
  • evals: give MCP scenarios a gateway so they can reach their tools (#563) (d9c629e)
  • people: read a contact's signature from their newest signed mail, again on new mail (#649) (d27ff6e)
  • people: time the contact keeper's next run from its last run, not a restart (#650) (787d070)

0.6.6 (2026-10-26)

Added

  • ui: a nonce Content-Security-Policy on every page (#623) (85b5945)
  • ui: confirm changes inside a frame once the browser reports the question seen (#635) (9d9ca16)
  • ui: let host origins frame the UI and call a short host list (#628) (42b626d)
  • ui: sum up Home under At a glance with more tiles (#633) (6d20d9c)

Fixed

  • briefing: stop first-week nudges for steps already done (#632) (c04bd42)
  • chat: stop the host chat list growing once the mark is stored (#634) (efe9dd4)
  • integrations: save briefs and digests into the chat-app conversation (#636) (e500883)
  • ui: count done optional rows and make adding the team optional (#631) (f211b96)
  • workflows: let chat fix a saved workflow on a card and say whose calendar a step reads (#629) (582449d)

0.6.5 (2026-10-24)

Added

  • delegation: let the own-mail hold follow the safety level (#614) (297218b)
  • orchestrator: set the safety level per channel and check all outside content (#612) (f0ae974)
  • ui: explain the safety levels with one worked example (#618) (fa5fb15)

Fixed

  • chat: stop false mail flags and the reply's note about them (#617) (7475f18)
  • delegation: prove the sender on Outlook searches too (#620) (c0077f6)
  • delegation: skip the content check on the Executive's own sent mail (#619) (6299964)
  • onboarding: suggest a website's team as people and show setup's progress (#610) (5ae4da4)
  • workflows: send scheduled runs as a short message and never a cut-off brief (#616) (a1964c8)

0.6.4 (2026-10-21)

Added

  • briefing: show the replies you're waiting on in Right now (#606) (823b230)
  • delegation: let a workflow answer an email as Handle it for me would, behind an injection check (#605) (653d80e)
  • delegation: send a chat draft written as you on your tap (#604) (211a484)
  • departments: let goals stand on their own with an owner (#594) (3efcb09)
  • onboarding: choose Personal or Team automatically and save goals with owners (#595) (b1f6a1c)
  • onboarding: draft setup from the company's website (#597) (af9f387)
  • workflows: let a workflow email as its creator, as a draft or sent (#600) (21e14e4)
  • workflows: let a workflow read its creator's mail and calendar (#601) (cf86c89)
  • workflows: let a workflow that reads mail send to recipients its creator chose (#602) (384a024)
  • workflows: let action steps use the Executive's chat tools (#599) (18449b7)

Fixed

  • briefing: skip a sent-mail thread that fails to open in waiting-on (#607) (5012783)
  • briefing: stop serving a Right now that a meeting has overtaken (#603) (f174588)

0.6.3 (2026-10-21)

Added

  • delegation: attach up to 15 Drive files to a draft (#581) (3558823)
  • memory: file each Learning miss under a finer cause (#590) (bcdc874)
  • observability: add opt-in Logfire request tracing (#584) (3acb995)
  • orchestrator: add activity, private-row and caller options to extension tools (#576) (7677540)
  • ui: split Learning by category and forget several at once (#588) (d4c84bc)

Fixed

  • briefing: read the owner's own calendar in the briefs too (#591) (57c059a)
  • briefing: stop calling sent drafts unsent and stop echoing the reader's chats (#585) (36f45dd)
  • briefing: write Right now from what moved today, with cited points (#589) (ce9b06b)
  • chat: tell the model the current date and time on every chat turn (#580) (1196957)
  • email: prove senders on domains without DMARC and on Outlook mailboxes (#583) (26a39ff)

0.6.2 (2026-10-20)

Added

  • briefing: show the live brief as a Right now card on Home (#566) (9121b20)
  • chat: approve held steps with a tap instead of saying go (#573) (0bb0d5a)
  • chat: let the Executive change things in the app from chat (#541) (f36818b)
  • chat: let the owner choose when outside content holds a turn (#571) (96c6510)
  • delegation: add Suggested actions On and a clearer training layout (#560) (0c96fd4)
  • delegation: answer meeting invites from a card, and on its own under Handle it for me (#568) (85f457e)
  • delegation: attach Drive files to drafts and let a go run held steps first (#555) (c887632)
  • delegation: let the Executive rename and move the company's own Drive files, with a tap card for the rest (#569) (7124221)
  • delegation: let the person choose sender only or everyone on a reply card (#556) (43561cf)
  • delegation: show reply cards closed with a one-line summary, open on tap (#570) (59ef8e0)
  • delegation: show What it does for you beside each Delegate tab (#561) (4648051)
  • integrations: approve cards with buttons in Telegram (#542) (62bde0b)
  • integrations: read Excel files in chat and images attached to emails (#537) (322659d)
  • integrations: transcribe voice notes and add a mic to web chat (#539) (b3233e0)
  • mcp_server: let an AI app's turn take actions when the request allows it (#533) (30319d9)
  • people: keep contacts up to date from sent mail and meetings (#557) (1aa8966)
  • people: reach people in the chat app they last talked in (#544) (45ad765)
  • people: remember what you deal with each person about and keep their contact details (#579) (f719738)
  • pick chat apps by name, describe watches, tap answers on replies (#543) (0baf9a1)
  • ui: add How it learns card and rename the screen to Learning (#550) (d1c50e5)
  • ui: keep the Topic view short with one-line topics and five per band (#538) (511e1a9)
  • ui: show each chat's app logo instead of a channel pill (#536) (3a9e2e9)

Fixed

  • briefing: call people in the live header by name, not their address (#572) (3f697e2)
  • briefing: make briefs report the reader's day, not their own messages (#545) (4b450ba)
  • briefing: name waiting drafts and say what approved app cards did (#558) (80bfc74)
  • briefing: treat the reader's own messages as background, not news (#548) (029a7ab)
  • chat: start each tool round's text on a new paragraph (#562) (dda3fc5)
  • delegation: draft a reply when a contact proposes a time (#559) (0413f78)
  • delegation: draft a reply when an answer to a call request leaves the time open (#554) (479baf6)
  • delegation: read the person's own last email before sorting a reply (#553) (9b594bb)
  • delegation: stop drafts writing the sender's name twice above the signature (#535) (c049abb)
  • delegation: work a typed answer into the reply draft instead of pasting it (#567) (a182d80)
  • integrations: reply to a voice note with just the answer (#547) (c14ff9e)
  • memory: let a name reach a lesson and keep the grader's proof to the person's words (#551) (528e546)
  • memory: review chats that read mail and count misses the product recorded (#549) (fbcf3d2)
  • scheduler: stand down a follow-up already handled in chat (#540) (4ac9af2)
  • ui: show last period's count instead of a signed change on Done for you (#552) (15964d9)

Changed

  • orchestrator: shrink the cached chat prefix (#546) (50a8f1b)

0.6.1 (2026-10-18)

Added

  • briefing: show a first-week card on Home instead of the getting-started brief (#523) (fd53d82)
  • chat: group chats by topic on the Chats page (#521) (3c6497e)
  • chat: keep a long topic list short with areas and quiet bands (#531) (88bbb78)
  • onboarding: let setup answers be changed and redrafted (#527) (cecb9f8)
  • ui: group the Settings pages under headings (#530) (86f73e5)
  • ui: say what each finished run did in Pulse's Done for you (#528) (4a4494b)

Fixed

  • alerts: answer a create_alert call missing its subject or body (#518) (3b6975c)
  • chat: open and delete Google Chat chats from the web (#526) (b3600f1)
  • chat: show the Executive's name over its replies (#524) (5b9b79d)
  • chat: word next-message holds as a yes with a reason (#532) (b62af21)
  • memory: keep an adapter's speaker check on turns the chat review grades (#520) (cb49225)
  • ui: make the getting-started Ask button open a chat, and drop "overnight" (#522) (95f98ef)
  • ui: show the Home ⋯ menu only in a chat (#525) (c29a90a)

0.6.0 (2026-10-16)

Added

  • chat: check a reply's claims against the tools it ran (#487) (dcde468)
  • chat: let people in a shared chat write in it live (#495) (fa4d503)
  • chat: review quiet chats and fix what the replies left undone (#490) (cd19d1b)
  • chat: show a meeting's approval card under the chat reply (#512) (5f9ad3c)
  • delegation: save a file from the owner's own mail to Drive or OneDrive (#499) (b0489b2)
  • delegation: show whether the mailbox sign-in includes the calendar (#492) (b0ffa85)
  • mcp: answer MCP clients as the signed-in caller (#498) (07ffd1e)
  • memory: grade each turn in the conversation review and show what it got wrong (#504) (8669a84)
  • memory: keep a lesson from a miss and read it on later turns (#505) (da44f43)
  • memory: replay the misses nightly and show how many would go right today (#506) (4c79fa4)
  • orchestrator: consult specialists for judgment questions (#484) (ad56027)
  • ui: give Handle it for me its own picker and split it from Act as me (#493) (18356a1)
  • ui: keep the How far it goes as you dial on Act as me only (#482) (bb8e4c5)
  • ui: let the help panel answer questions about any part of the app (#494) (5f734e3)
  • ui: show each Delegate option as steps and rename For my inbox too (#514) (15734cc)
  • ui: show plain contact lines on a person's page and name them in the breadcrumb (#481) (8c8da3a)
  • ui: show who writes and who sends on What it may do (#496) (3f1d22f)

Fixed

  • alerts: label alerts team members raise and limit what they can dismiss (47084a3)
  • briefing: date inbound mail, name the reader and stop repeats in the brief (#485) (871d0a6)
  • briefing: label the principal's own chat messages in the brief (#515) (1888e3c)
  • briefing: stop counting chat drafts sent from Gmail as waiting (#486) (23a76eb)
  • briefing: write every brief to its reader as you, never by name (#516) (e9e4309)
  • calendar: read a booked event's id and Meet link only from the reply's own link (#511) (a65de76)
  • calendar: read workspace-mcp's plain-text reply when booking or cancelling (#510) (635de4a)
  • chat: count a channel whose adapter verified the sender as a verified surface (#513) (0dd25b1)
  • chat: find the JSON object around prose in fast-model answers (#488) (df9d7a0)
  • chat: stop teammates' lines in a shared chat from steering sends (#501) (229c119)
  • delegation: forward a chosen message and stop the second mail-read refusal (#491) (f6d61bc)
  • delegation: stop false refusals and stray folders when saving mail files (#502) (6fe4c63)
  • integrations: hold the whole turn for mail from outside People (#508) (c200368)
  • monitoring: check every redirect hop and pin the fetched IP (#507) (49e0872)
  • orchestrator: hold writes and new contacts after a turn reads outside content (#503) (58f9aa2)
  • scheduler: run a teammate's scheduled follow-up as that teammate (#500) (1ee5fd6)
  • ui: say plainly what Take the lead On skips (#497) (a77d55b)

Changed

  • config: build settings once and rebuild only when their inputs change (#489) (04612b1)

0.5.6 (2026-10-14)

Added

  • briefing: approve a proposal card on the server (#458) (99eca81)
  • briefing: open the first week's briefs with getting-started moves (#465) (1a28256)
  • chat: pin, rename, share, export and redo the last turn (#471) (8068e43)
  • delegation: add What it may do and one dial for acting as you (#473) (c8eec03)
  • delegation: read the person's own calendar under Act as me (#469) (b94352b)
  • onboarding: open the first hour on a question and the first week's brief (#476) (6a232ec)
  • orchestrator: hold a chat-channel sender's sends when they match no one on People (#462) (5541ed0)
  • orchestrator: put a team member's chat sends and edits under department authority (#459) (8b195c2)
  • today: show the owner what the Executive did and tell people their tasks (#464) (de3c788)
  • ui: redraw the Executive's mark as the Council Ring (#461) (35808c2)

Fixed

  • chat: re-read a resumed chat's transcript every turn (#468) (f9fcb8b)
  • delegation: resolve the Act as me caller through the shared caller resolver (#478) (7e04496)
  • delegation: resolve the What it may do viewer through the shared caller resolver (#477) (7189ce5)
  • delegation: show all-day calendar events on their own day (#470) (00290b4)
  • integrations: give email turns, gateway calls and the chat stream deadlines (#463) (a7ef7a9)
  • integrations: log a mail source whose loop stops with an error (#474) (5bf5dba)
  • integrations: put an email that got no reply in Needs you (#479) (4b6ede6)
  • orchestrator: find mail the reader is on even when a teammate is too (#475) (25c827b)

0.5.5 (2026-10-13)

Added

  • api: answer a chat to the end on the server with POST /chat/background (#431) (e8f2b19)
  • api: report the pinned Google tools priming could not find (#428) (b49c5f6)
  • chat: give each chat in the list its last line (#440) (d9d2d55)
  • departments: check in weekly by default, with a plain schedule picker (#455) (861c8dc)
  • evals: pick the judge model with EVAL_JUDGE_MODEL (#446) (0964522)
  • integrations: say it is an AI on mail and chat the Executive sends (#449) (b5677f4)
  • onboarding: send the owner to setup, keep the interview, name their row, add a checklist (26451db)
  • people: add quiet hours and a digest setting for the Executive's own messages (#457) (0545f35)
  • scheduler: retry the MCP gateway and say why a brief was not sent (#448) (d11d802)
  • slack: take Slack events over HTTP at /webhook/slack (#444) (9a662f2)

Fixed

  • alerts: name chat senders by their People record in triage (#434) (85f0350)
  • api: drop the localhost CORS origins on a public deployment (#452) (e8bac77)
  • audit: log model usage from every call site (#427) (37d97b3)
  • briefing: let owners snooze cards and rerun failed scheduled runs (#456) (714ade9)
  • chat: keep a teammate's chats and activity private to them (#439) (4d39810)
  • chat: keep a teammate's mail with the Executive private to them (#433) (80c297b)
  • chat: word the suggested follow-up as the user's own request (#442) (b0bad78)
  • integrations: drop replayed and retried Slack events (#454) (ef390b7)
  • integrations: keep outside text from booking or cancelling meetings (#443) (290f6d3)
  • integrations: retry a Telegram reply as plain text only on a refusal (#437) (681e237)
  • integrations: show Telegram replies' Markdown as formatting (#436) (e510133)
  • knowledge: read scanned PDFs from Drive with OCR (#453) (eb70dfe)
  • providers: keep the question's shape so the second call reads its cache (#445) (0f55f64)

Changed

  • chat: cache each question's context once and cap knowledge excerpts (#432) (bc6d4e8)
  • chat: read only the start of each chat's last message for its preview (#441) (fa86287)

0.5.4 (2026-10-10)

Added

  • chat: cut chat-app chats into conversations and search inside chats (#413) (3915b6b)
  • chat: let a later-registered adapter cut its own old chat history (#416) (07d6af5)
  • delegation: edit a waiting reply on its card and keep replies out of Drafts by default (#423) (c480289)
  • org: start new workspaces with no departments (#422) (e2574c1)
  • ui: lay out Pulse as one overview with focused screens (#418) (62f1db4)
  • ui: put Act as me, Handle it for me and Take the lead under Delegate (#417) (faf20f2)
  • ui: show what Pulse remembers above the schedule lists (#421) (d20526f)
  • workflows: fold playbooks into workflows (#415) (f41a915)
  • workflows: let an extension offer its tools to workflow action steps (#419) (d765eef)
  • workflows: run a job in the local sandbox when the remote runner turns it away (#424) (ff43129)

Fixed

  • prompts: scope the Google manifest to the services and tier workspace-mcp loaded (#425) (be279b5)
  • ui: tidy the new-workflow chat and the page after saving a workflow (#412) (a13d6f3)

0.5.3 (2026-10-09)

Added

  • agents: refresh the default models (#390) (b747bee)
  • chat: name connected tools in plain words and collapse repeated chips (#394) (1ce8d97)
  • delegation: add an In training step to Handle it for me (#403) (f581e8c)
  • delegation: add reminders and forward-as-draft to Act as me (#387) (bbfc6c3)
  • delegation: find and reopen the person's own emails across conversations (#397) (49ed3cc)
  • delegation: leave approval cards for actions an email calls for (#388) (44a1807)
  • delegation: lock down only the turn that read the mail (#385) (a794917)
  • delegation: put each Act as me setting in training on its own (#408) (7ae0737)
  • delegation: read the speaker's own mailbox and attachments from chat (#384) (5ae7674)
  • memory: let each person share their work style with the team (#391) (dee0275)
  • orchestrator: add an In training mode to Take the lead (#400) (57937ac)
  • orchestrator: let an installed extension add chat tools and a kind of document (#410) (dc3c105)
  • scripts: let the Executive build and keep its own tools (#383) (5e4348d)
  • workflows: edit a saved workflow by conversation (#380) (a4677ad)
  • workflows: let Python jobs run on a separate runner (#402) (d266972)
  • workflows: run Python jobs on files and keep them as custom tools (#386) (f7a759d)

Fixed

  • chat: label a calendar delete through manage_event as a removal (#395) (f060ee7)
  • delegation: ask instead of block after reading mail (#405) (4a2ff41)
  • delegation: let Drive, Docs and Sheets reads run after a mail read (#396) (643287c)
  • delegation: lock only links, scripts and broadcasts after a mail read (#401) (deb44bd)
  • deps: bake the tool-search model with the fastembed the gateway runs (#392) (96d5e3a)
  • orchestrator: gate send_slack_dm on the People roster (#404) (14f785f)
  • orchestrator: send a message only to the principal without an approval card (#393) (417771c)
  • orchestrator: withhold run_script from turns that are not the principal's (#411) (f87f137)
  • workflows: offer Python jobs without files and name a runner's extra libraries (#409) (e290c37)

Changed

  • orchestrator: cache the conversation history between messages (#389) (2f9560b)
  • orchestrator: offer less common tools on demand and trim tool results and specialist lookups (#406) (303d45e)
  • orchestrator: share one reading budget per turn and build a tool for multi-read answers (#407) (1548742)

0.5.2 (2026-10-06)

Added

  • delegation: let people describe how they write in their own words (#362) (593741e)
  • onboarding: describe your work first, then ask what's missing (#377) (8e112dc)
  • providers: default to Claude Sonnet 5.5 and Opus 5.5 (#371) (12b89ce)
  • ui: move your own memory card to Settings as About you (#359) (dad7384)
  • ui: play the Council Consult animation while the Executive works (#368) (639f968)
  • ui: reword the who-is-it-for setup choice as Personal and Team (#375) (892e96b)
  • workflows: let the morning reflection search the knowledge base (#364) (e7106ab)

Fixed

  • delegation: let the email's own greeting decide who a group email asks (#365) (8145855)
  • integrations: stop logging the start of each email body (#367) (35a1e6c)
  • onboarding: keep going when a setup turn returns prose or a null hint (#378) (fd3e7bb)
  • people: point roster emails at Home and name who a reply is about (#366) (8c15b9d)
  • ui: hide the Always asks first switches while Take the lead is off (#360) (f1988d6)
  • ui: keep the watch list page from scrolling the whole screen on phones (#363) (4cd1136)
  • ui: put Dismiss and Edit in Gmail beside Send on reply cards (#358) (7a278fa)

0.5.1 (2026-10-04)

Added

  • delegation: send inbox replies on their own under Handle it for me (#353) (23af959)
  • orchestrator: let the Executive take the lead behind one approval gate (#355) (f341051)

Fixed

  • ui: harden the UI for phones, readability and plain wording (#357) (d46eee5)

0.5.0 (2026-10-03)

Added

  • chat: fold messages sent mid-turn into the running answer (#352) (bba990f)
  • delegation: check Act as me drafts against what the writer already told those people (#350) (0543e92)
  • memory: bring Always in the loop notes into the owner's briefs and remind when due (#348) (a5b399d)
  • memory: keep notes of what people say in chat (#346) (eebda07)
  • memory: keep private notes from replies you approve and send (#342) (96098b7)
  • ui: add the History tab and settings for Always in the loop (#344) (870ff2b)
  • ui: move the Agent Council's Advanced toggle into the agent panel (#345) (29de6ca)
  • ui: simplify every screen with a six-place menu and focused pages (3155921)

Fixed

  • artifacts: keep each document private to the person who made it (#351) (cbaad6f)
  • delegation: stop drafting inbox replies to group emails that ask someone else (#343) (5da0b29)

Changed

  • ci: run tests with temp files in RAM and parse the package once (#339) (75e3163)

0.4.6 (2026-10-02)

Added

  • auth: add SSO sign-in through any OpenID Connect provider (#338) (d338f1f)
  • delegation: let Act as me use an Outlook mailbox (#336) (6b7957e)
  • integrations: add a Microsoft 365 (Outlook) mail and calendar backend (#326) (5ec59a4)
  • integrations: sync OneDrive folders into knowledge and add OneDrive tools in chat (#331) (29d3037)
  • knowledge: sync Confluence spaces into the knowledge base (#324) (27b44bc)
  • settings: show the running version and when a newer release is out (#323) (5d23f29)

Fixed

  • ui: make npm run lint real again, package-wide, and enforce it in CI (#327) (398519f)
  • ui: make the web app readable and usable on phones (#335) (3ff5f50)

0.4.5 (2026-10-01)

Added

  • agents: add additional instructions to Agent Council agents (#308) (4bc9567)
  • agents: add Fast, Balanced and Thorough quality presets to the Agent Council (#309) (58ba2ec)
  • agents: offer Direct, Supportive and Analytical voices in setup and Settings (#310) (c99d9d1)
  • agents: open the Agent Council in a simple view (#311) (4e45588)
  • delegation: draft replies to your inbox and send them on your tap (#287) (0980b75)
  • delegation: let team members use Act as me, with their mail private to them (#320) (4b370b0)
  • knowledge: put company documents first and list Drive and Notion files (#306) (7f1acd8)

Fixed

  • knowledge: accept Excel and CSV uploads on the documents page (#318) (6ae0e6b), closes #316
  • knowledge: start one sync per double click and report partial sync failures (#307) (649cdd8)
  • slack: show senders when a thread message goes unanswered (#319) (6293b6f)
  • ui: pass uploads up to the api's limit through the middleware (#303) (0977d06)

Changed

  • knowledge: parse uploaded documents in a short-lived child process (#305) (cd8b3ec)

0.4.4 (2026-09-30)

Added

  • alerts: let the principal find and clear alerts off the live board (#299) (7bee3dc)
  • attunement: let people assign a task to a teammate (#285) (13da433)
  • knowledge: add the files in chosen Google Drive folders to the knowledge base (#279) (354ddb4)
  • memory: let teammates record attributed standing facts (#280) (ede7b86)

Fixed

  • deps: move the mcp gateway's dependency cutoff past fastembed 0.8.0 (#288) (8098ed5)
  • integrations: acknowledge a new email sender only when Gmail authenticated them (#283) (35e00d3)
  • integrations: pin the mcp gateway to extensible-mcp with a leaner tool index (#291) (1a49047)
  • integrations: pin the mcp gateway to extensible-mcp with batched embedding (#290) (5b342bc)
  • knowledge: share one lean embedding session across chroma collections (#292) (b05a75b)
  • mcp: pin extensible-mcp to the commit that supports mcp 2.x (#298) (d06bae2)
  • memory: keep strangers' email out of decisions under one untrusted-content policy (#284) (966821a)

0.4.3 (2026-09-29)

Added

  • briefing: feed the principal's live day into the header and the brief (#263) (7fea2d5)
  • briefing: ground names and figures in unattended prose before delivery (#272) (a565c30)
  • memory: keep chat corrections as standing facts in prompts (#270) (4e2a42b)
  • memory: remember the Drive files a conversation found or read (#276) (f200cd9)
  • orchestrator: tell the executive which systems are connected (#271) (8c742b3)
  • people: let the owner confirm and add unknown senders (#273) (ff693ae)
  • providers: add LOCAL_REASONING_EFFORT for thinking-only models (#274) (ed06025)

Fixed

  • ui: show turn progress while the executive is still working (#261) (29f634b)
  • update the Apache license file (#275) (421db3c)

0.4.2 (2026-09-28)

Added

  • delegation: draft replies in your voice from your own Gmail (#242) (cc2e95f)
  • knowledge: read scanned PDFs on every channel (#254) (a083f18)
  • providers: read PDFs through each provider's own file support (#257) (9e0bbc4)
  • ui: give the settings page a section nav and grouped tools (#260) (f896562)
  • ui: make goals quick to add with one required field (#239) (5219d0a)
  • ui: open workflows on a start-here panel (#236) (ab37898)

Fixed

  • delegation: show your whole writing profile and keep its line breaks (#252) (3f6ba5b)
  • evals: point make eval and docs at the packaged scenarios (#240) (a0f31d9)
  • integrations: record the inbound email's text in its audit row (#259) (8d1ca4b)
  • orchestrator: close the paths around the Google Workspace egress gates (#258) (eb230cc)
  • ui: anchor the auth middleware's path exclusions (#251) (b90c611)

0.4.1 (2026-09-25)

Added

  • add local login for make dev and link the owner at setup (#218) (a8efeed)
  • agents: add a sales specialist and advice for bootstrapped founders (#220) (fd7a8aa)
  • briefing: email briefs as a backup and show when one isn't sent (#226) (71a75c8)
  • chat: keep answers when an expert fails and show their sources (#222) (90ca2b2)
  • core: add workspace settings for your time zone and a solo mode (#221) (345a91a)
  • goals: track your goals and dated commitments in solo mode (#228) (242af54)
  • onboarding: ask what your role is when you use Open Executive just for yourself (#230) (5dc3a43)
  • orchestrator: work for one person in solo mode (#225) (1c35275)
  • people: keep contacts apart from your team and act on mail you forward (#233) (1bf703d)
  • ui: add a setup status page that tests each part of the install (#219) (a95f924)
  • ui: describe Just me for any role, not only someone running a business (#234) (7d97b8d)
  • ui: set up and run Open Executive just for yourself (#224) (bf3670a)
  • workflows: add a weekly review and a daily top three (#232) (62b283f)

Fixed

  • hold escalations, allow cancel, and keep roster edits owner-only (#216) (4975dff)
  • people: stop teammates taking over the owner's entry (#227) (9325113)
  • scheduler: run check-ins, chase idle initiatives and honour brief channels (#223) (2a31596)

0.4.0 (2026-09-24)

Added

  • artifacts: deliver artifacts by message link or email attachment (#209) (212cfaf)
  • artifacts: publish, reread and revise artifacts in html, docx, xlsx or links (#207) (066abd3)
  • knowledge: make chat-proposed playbooks drafts a person approves (#208) (b775dea)
  • scheduler: add a global pause for the executive's autonomous work (812a39e)
  • ui: add a vitals panel to the Pulse heartbeat card (#203) (4101207)
  • ui: manage skills as playbooks on the Workflows page (#202) (4691e48)
  • ui: rename Artifacts to Documents for users and the Executive (#214) (67ee80d)
  • ui: share one sidebar with a short recent-chats list on every page (#195) (59b7d35)
  • ui: show every peer-memory note in a scrollable People tab (#201) (daabb99)
  • workflows: confirm a workflow's first write to a new target (#199) (d63d11d)
  • workflows: have workflows follow playbooks (#206) (1c61e18)
  • workflows: let workflow steps use tools (#193) (429aeda)

Fixed

  • api: check session ownership on every per-session route (#197) (cf61609)
  • api: remove the unauthenticated /debug/last-turn endpoint (#200) (112486d)
  • ui: call the Jobs page "Workflows" (#194) (d3e48cb)
  • ui: compare complete days only in the heartbeat trend (#204) (f4a68b6)

Changed

  • core: parallelize unit tests with pytest-xdist (#213) (8b9a822)

Documentation

  • release: document forcing a version with Release-As (e71a025)

0.3.2 (2026-09-23)

Added

  • ui: add an assistant-led workflow wizard and compact the jobs catalog (#189) (c52b96c)

Fixed

  • ui: validate check-in cadences and clear emptied leave dates (#191) (5940557)

0.3.1 (2026-09-23)

Added

  • chat: suggest a follow-up in the composer (#182) (7f7c1c1)
  • ui: group council model picker by provider (#186) (8b3fb2f)

Fixed

  • memory: quote extraction and open loops from the speaker's own words (#187) (100f7d9)
  • memory: record only the person's own words in peer memory (#185) (c32cf49)

0.3.0 - 2026-09-23

Added

  • Attunement: the Executive follows up on what people owe (#175). When anyone on the roster commits to something in chat ("I'll send the vendor quote Thursday"), asks for something, or the principal says a teammate will do something, it becomes an open loop. Loops come due, get chased by the nudge engine through the usual routing and outbound checks, and close when their owner says it's done. The Executive can list them ("what is Sara waiting on?") and close one on request. Each person's page lists their open loops with Mark done. Every chat message now records who actually sent it, so nothing an outsider writes is read as someone on the roster.
  • 👍/👎 on replies (#175), in the main chat and the Ask OE panel.
  • Attunement: learning which proactive messages land (#176). Every proactive DM (nudges, follow-ups, reflection, research and alert-review messages) is resolved as replied, acted on, ignored after 72 hours, or void when it stopped mattering. Credit only goes to the person who acted. A kind of nudge a person's last five resolved sends all went unanswered on ranks last for them on a longer cooldown until they answer one. The morning reflection gets a "What lands" summary, and each person's page a "How they respond" card.
  • Attunement: per-person working style (#177). Up to four short rules on how to write replies for each person ("lead with the recommendation, then the numbers"), learned only from their own messages and 👍/👎 and pinned into their own conversations. Rules are checked before they are stored and every time they are used, and must be about how replies read, never an action. A "How I work with them" card lets the principal or the person edit, lock or reset them.
  • New settings, all with defaults: ATTUNEMENT_* in .env.example. ATTUNEMENT_ENABLED=false turns off open-loop tracking and style learning; the outcome ledger has no switch and always records. None of it needs Honcho.

0.2.2 - 2026-09-22

Changed

  • The MCP gateway runs a pinned extensible-mcp commit. It was launched from the repo's default branch, so every container start ran whatever that branch held that day: a versioned image did not pin its gateway, rolling back an image did not roll the gateway back, and a start without network failed. The gateway and the image's pre-warm now launch the same commit with the same --exclude-newer cutoff, which also freezes extensible-mcp's own dependencies (uvx re-resolved those against PyPI on every start), and a unit test keeps the two in step. An image whose pre-warm succeeded now starts the gateway without network; the pre-warm is still best-effort, so a build during a GitHub outage ships an image that fetches it at first start. Updating the gateway is a deliberate bump of the commit and cutoff.

Fixed

  • The compose health check no longer fails on a healthy API. It runs curl -f http://localhost:8000/health inside the API container, but the image never installed curl, so the check failed on every run and compose reported a working API as unhealthy. The API image now includes curl.
  • The People tab shows the newest notes about a person (#174). Its "recent" list and "learned" date asked Honcho for conclusions with reverse=True, which returns the oldest page, so a person's correction made in chat never appeared there. It now reads the newest page.

0.2.1 - 2026-09-22

Changed

  • The API image installs the CPU-only build of torch. torch is only present because sentence-transformers needs it, and the container runs on CPU hosts, but the default Linux wheel is the CUDA build and pulled in nineteen packages the image never used — fifteen nvidia-* libraries, three cuda-* shims and triton, about 2.2 GB of compressed wheels. torch now resolves from PyTorch's CPU index (2.13.0 → 2.14.0+cpu on Linux, plain 2.14.0 on macOS, both from that index) and the Dockerfile installs from the lock with uv sync instead of an exported requirements file, so each package comes from the index the lock names. Expect the API image to shrink by several GB and cold builds and pulls to get much faster. uv sync gets the same wheels locally; a GPU deployment would need to override the index, and a plain pip install of the package (which ignores uv sources) still gets the CUDA build.

0.2.0 - 2026-09-22

Security

  • The sign-in allow-list is now the union of ALLOWED_EMAILS and the People roster, not the roster alone (#132). The UI previously treated the roster as authoritative as soon as it held one email, so any Person row with an address silently disabled ALLOWED_EMAILS — a fixture load, which wipes people and inserts its own addresses, could lock the configured operator out of their own instance. This widens access on upgrade: every address still sitting in ALLOWED_EMAILS regains sign-in even if that person is not on the roster, so audit the env var before deploying. Removing someone now means removing them from both the roster and ALLOWED_EMAILS. The authorized callback's fail-open also narrows: it used to admit any valid session whenever the roster fetch failed, and now admits only a session that is not in ALLOWED_EMAILS and whose roster membership is unreadable.
  • packages/ui: force lodash-es to 4.18.1 via an npm overrides entry (GHSA-r5fr-rjxr-66jc code injection in _.template, GHSA-f23m-r3pf-42rh prototype pollution in _.unset/_.omit). The vulnerable 4.17.23 was pinned exactly by [email protected] underneath [email protected], and no mermaid release moves off it.

Removed

  • Talent / executive search. The whole vertical is gone: the talent specialist, openexecutive/talent/ (engagements, candidates, offers, the ChromaDB matching graph), its 21 REST routes, 12 chat tools, 4 MCP tools and the /talent UI, plus the five recruiting workflows (candidate_screen, candidate_outreach, interview_coordination, reference_check, offer_approval). The chro specialist and its comp_refresh / org_design / performance_review / exec_search_brief workflows are unchanged — exec_search_brief is an advisory hiring brief, not pipeline.

  • Staff onboarding. openexecutive/staff_onboarding/ (templates, per-hire plans, tasks), its REST routes and 8 chat tools, the new_hire_onboarding and role_onboarding workflows, the /staff-onboarding UI, and the onboarding_ramp / onboarding_kickoff / onboarding_checkin scheduler kinds. The company-setup wizard (/onboard) is untouched — it is a different subsystem that happens to share the word.

  • GET /today no longer returns the talent or onboarding fields, and the per-turn chat briefing no longer includes their digests.

    Existing SQLite tables (engagements, candidates, offers, onboarding_templates, onboarding_plans, onboarding_tasks) are not dropped — nothing reads them, and a blank client slot still wipes them so candidate data cannot cross slots. Pending talent reminders the removed workflows had scheduled on the principal's DM channel are cancelled by a one-time startup sweep (cancel_orphaned_talent_reminders, recorded in the new app_migrations table); the sweep can be deleted in the release after next.

Changed

  • The chat turn ceiling is now 300s (360s with Committee review), up from 120s/180s: deep multi-specialist turns were being cut off mid-answer. A ceiling this generous is only reasonable because a turn can now be ended by the user, so the two changes ship together. The onboarding interview, which previously borrowed CHAT_STREAM_TIMEOUT_S, gets its own INTERVIEW_TIMEOUT_S (still 120s) — it retries twice, so inheriting the new ceiling would have meant a 10-minute hang before the wizard surfaced a timeout.

Fixed

  • A turn broken off early is no longer missing from the next turn's context. On the disconnect and timeout paths the route persisted the partial turn to SQLite, but the Executive's post-turn block — which mirrors it into the live in-memory session — was skipped, and a cached session never re-reads its history from the DB. The next turn in the same process then prompted as though the turn had never happened, while a page reload showed it. The route now mirrors the turn itself on every broken-out path.
  • The UI proxy now forwards client disconnects upstream. signal: req.signal was missing from the backend proxy's fetch, so the API never saw http.disconnect and its request.is_disconnected() check could not fire in production: closing a tab left the turn running to completion against Anthropic and the partial reply was never saved.
  • A disconnected turn no longer keeps working after the client is gone. The SSE driver races the stop switch with asyncio.wait, which — unlike the asyncio.wait_for it replaced — does not cancel its futures when the task awaiting it is cancelled. Since Starlette cancels the response body on http.disconnect, the in-flight step is now cancelled in a finally, so a closed tab cannot leave a specialist or tool round running with no deadline and no persistence.

Added

  • Versioned container images on GitHub Container Registry (#142). Every push to main publishes ghcr.io/sentelabsai/openexecutive-api:main and …/openexecutive-ui:main; pushing a vX.Y.Z git tag publishes X.Y.Z, X.Y and latest. Deployments can pull a pinned version instead of building from source. See docs/deployment.md → Images.

  • Stop button in chat. A reply can now be halted mid-stream, from the main chat composer and the Ask OE side panel (Escape works too). Whatever the Executive had written is kept, persisted and marked Stopped by you, so a truncated answer is not read back as a complete one after a reload. The client mints a client_turn_id and sends it with the turn; POST /chat/stop halts it. Keying on a client-minted id is what makes the button live from the moment Send is pressed — the server spends several seconds fetching context before the response stream exists, and a stop inside that window costs nothing because no model call has been made yet. An unknown id and another caller's turn both return 404, so the endpoint cannot be used to probe which turns are live.

  • Conversational onboarding. /onboard now opens with "tell me about your company" instead of a 12-step form. The user writes a paragraph (and can attach a deck, one-pager or brief), the new onboarding_interviewer agent asks up to 8 clarifying questions, then drafts a company profile, leadership roster and department list that the user edits inline and saves. The draft review reuses the /company-profile section editors, extracted to components/company-profile/ProfileSections.tsx, so the editing surface is the same one the user gets permanently afterwards. New endpoints: POST /onboard/interview/{start,message,draft,commit} and GET /onboard/interview/{session_id}.

    Design notes: the interview never writes — POST /onboard/interview/commit is the single write, ordered so a rejected save leaves the draft editable and retryable. Departments are reconciled additively (matched ones updated, new ones created, none ever deleted), unlike the fixture loader which wipes the table and would destroy the eight defaults' specialist_key wiring. Exactly one person is marked principal and gets WILDCARD authority; no contact details are imported. Every failure path is asserted not to echo the user's input, because these transcripts carry ARR, burn and runway. The step-by-step wizard remains at /onboard?mode=form and still backs the openexecutive onboard CLI.

  • Research watch policy grounds in departments and recent decisions, and routes to department heads. Departments gain a watched_entities list (PATCH /departments/{slug}, edited one per line on the department page). Named there, an entity is strong grounding for the research watch policy — like a profile vendor or ticker — so a proposal about it can be added on its own, and the watch is inserted with route_to_department / route_to_person_id (the head). Those columns now actually route: every watch alert carries the department's head and a department:<slug> tag, so it queues on the head's briefing instead of the principal's. Charter scope phrases and goal key results ground suggestions only. Suggestions in a department's area go to its head as one "Watch suggestions for " card per run (refreshed in place, re-issued weekly after it is handled; the principal when the department has no head), and the principal's pile-up nudge counts only their own. The ten most recent episodic decisions from the last 90 days are rendered to the research council and add a point to a proposal whose entity they name (with the decision's department as a routing hint). The research fingerprint tracks department watch interests and the decisions that name a known entity, so a new watched entity or a relevant decision triggers the next scan. /watchlist shows "for: " on routed rows.

  • Watch proposals are linked to their evidence, and profile entries ground by name. propose_watch now requires finding_index, and when the model omits it the policy links the finding that cites the source itself instead of rejecting the proposal for lack of evidence. Profile competitor / vendor / ticker entries and department watched entities are parsed to their names ("Tesla (TSLA) — Model Y…" grounds as Tesla plus the ticker TSLA; "GM / Chevrolet (…)" as both), so short names such as BYD, GM or Kia match and the description text can no longer stand in for the entity.

  • Research watchlist policy — grounded watches go straight in, uncertain ones become suggestions. The research council's watchlist pass no longer adds watches itself; its only tool is propose_watch, and deterministic policy (monitoring.research.watch_policy) decides from company data. A proposal tied to a named competitor, vendor, ticker, initiative or priority (the profile gains vendors and tickers for this) and corroborated (own source, high-confidence finding, consensus, the policy's own track record) is added on its own — quietly: daily cadence, medium severity floor, a keyword trigger for feeds, 5 % for stocks — and shows up in the brief's "handled overnight" block as watching. Anything the Executive is not sure about lands as a dry-run suggestion (polls, never alerts) in a "Suggested by the Executive" section on /watchlist with Approve / Decline (reason: not relevant, too noisy, wrong source); the morning brief mentions the pending count in one line, and a single nudge alert fires only once ≥3 suggestions have waited ≥7 days. Declines are remembered by target and enforced in the tool handler, so a declined source is never re-proposed under a new slug; "Stop watching…" on a research-added card, remove_watchlist_entry and DELETE /watchlist/{slug}?reason= on research rows record one too. Approvals, declines, expiries and auto-disables feed the policy's history and the research turn ("stock watches grounded in a competitor: 4 approved / 1 declined"). Research-added watches retire themselves: the scheduler sweep disables one that fired ≥5 alerts with ≥2 dismissed and low trust, or one with 3 consecutive poll failures (audited; re-enable from /watchlist; silence alone never retires a watch). Endpoints: POST /watchlist/{slug}/approve, POST /watchlist/{slug}/decline. Settings: WATCHLIST_RESEARCH_MAX_DIRECT_ADDS=2, WATCHLIST_RESEARCH_MAX_PROPOSALS=2, WATCHLIST_MAX_ENABLED=40, WATCHLIST_PROPOSAL_TTL_DAYS=14.

  • Self-maintaining alert feed. Alerts get a per-category time-to-live (ALERT_TTL_DAYS_ACTION=14, ALERT_TTL_DAYS_MONITORING=3); a scheduler sweep expires past-TTL rows (audited, reversible via POST /alerts/{id}/reopen) and every surface reads the same live view. A repeat of an open alert with the same (source, dedup_key) now coalesces into the existing card (occurrence_count, last_seen_at) instead of stacking, and only re-pings when severity rose to high/urgent; monitoring alerts carry a stable watch:<slug> key so one watch means one card. A distinct resolved status keeps Executive closes separate from ack (user approved).

  • Executive alert review (alert_review_scan, every 6 h and right before the morning brief): re-examines each open alert with evidence — newer signals from the same watch, related alerts, activity since, the roster with SLAs, department authority — and, through deterministic policy, routes + DMs the owner (propose_via_alert for propose-only departments), nudges, escalates to the principal with a deadline, drafts an artifact, suggests a workflow, folds duplicates, or resolves with evidence (high confidence only, citing a server-minted evidence ref — free text never closes a card; otherwise annotates "likely stale"). Alert text is rendered as inert data inside the review prompt (injection boundary); board / comp / legal matters only ever go to a scope-holder or the principal and keep their text; route/escalate are idempotent across passes; DMs carry an "[Alert review] Re: …" header. Capped per pass (ALERT_REVIEW_MAX_MOVES_PER_SCAN), single-flight, off for every caller with ALERT_REVIEW_ENABLED=false, every move audited with its evidence and prior state, every close reversible. POST /alerts/review runs it on demand.

  • Briefing cards read as decisions: what changed since you last looked, the recommended move as the primary control, why-now / due chips, provenance (age, seen ×N), "N folded in", a "Handled overnight" rail with Undo, an "Executive handled N overnight" pill, a capped "Needs you" list with Show more, "Dismiss older than 7 days" (explicit ids), "Re-check relevance", and "Mute this topic" on dismiss.

  • POST /alerts/bulk-ack, POST /alerts/{id}/reopen, POST /alerts/review.

  • Dismissing a watch-sourced alert lowers that watch's trust_score (and bumps dismiss_count); approving recovers it. Trust now discounts the ranking and is shown to the review as evidence.

Changed

  • "Handled overnight" now says what the Executive actually did, and says it with some pride. The briefing rail no longer lists the review's changed verdict (an "Updated '' — …" row): that rewrite leaves the alert open in "Needs you", where the card already shows the note, so the row double-reported it and called it done. Rewritten items now render in the morning brief and end-of-day digest as a REWRITTEN block under "what changed" (brief_state.rewritten_since), and a rewrite alone still un-suppresses the brief. Each remaining /today.handled_overnight row is structured (headline, target, detail, outcome, evidence_ref, event_type, and the alert's status now) instead of one truncated sentence; routed / nudged audit summaries name the person rather than "person 12", and the nudge bookkeeping marker is stripped for display. The rail is now one line by default — a count-only first-person heading ("Since your last brief: N off your plate, N in others' hands, N waiting on you", plus drafts / suggestions / watch changes / other moves / back on your plate when present) — with the rows behind a "details" disclosure, and nothing at all on a night with nothing handled. The grouping and summary logic lives in packages/ui/src/lib/handled.ts and is covered by npm test. Expanded, rows read first-person ("Resolved X — evidence", "Dismissed X as stale — …", "Handed X to Dana Kim", "Chased Dana Kim on X"), escalations first, one row per alert, every row linking its evidence to a pre-filtered /audit (the audit page now reads event_type / q from the query string) and a still-open row jumping to its card. Undo is only offered while the close still stands; a close already undone shows "Reopened" after a reload.
  • Specialists are told what departments watch. The research context every specialist receives now carries a DEPARTMENT WATCH INTERESTS block (each department's watched entities), and the grounding rule admits a department interest or a company named in a recent decision. A Finance head who lists Brex now gets Brex findings; before, the specialists were never told and their grounding rule dropped them. The per-specialist finding cap is enforced in the tool schema and the parser, not only in prompt text.
  • The research run has its own controls. RESEARCH_WEB_SEARCH_MAX_USES (default 3) caps searches per specialist independently of the chat knob; RESEARCH_SPECIALISTS picks which of the seven specialists run; the routing pass can no longer start a workflow (run_workflow is withheld, suggest_workflow remains); the periodic run's default interval moves from 120 to 360 minutes.
  • Web search reaches every OpenRouter model. The provider feature gate stripped the web-search tool for non-Claude models, so a research specialist or the Executive pinned to Gemini, GPT, Llama, DeepSeek or Grok ran without search (and the research prompt then told it to emit nothing). The tool is now kept for any OpenRouter model and replaced by OpenRouter's own openrouter:web_search server tool (the deprecated web plugin is no longer used); only a self-hosted OpenAI-compatible backend still strips it. The server tool carries the same contract as Anthropic's: the model decides when to search, max_uses caps searches server-side, the domain allow/block lists apply, and the search count is reported, so RESEARCH_WEB_SEARCH_MAX_USES and WEB_SEARCH_MAX_USES mean the same thing on both deployment types.
  • Per-run watch budgets go to the strongest proposals. The policy classifies every proposal first and applies them by tier (direct adds before suggestions) and descending score, so when the model files more candidates than the budgets or the ceiling allow, a better-grounded proposal filed later no longer loses its slot to weaker ones filed earlier, and a duplicate target keeps its strongest filing.
  • Every model call is recorded. Research specialists, the research routing and watchlist passes, triage and the chat memory extractor now write the same cache_event audit row the Executive's chat turns always did (tokens, cache hits, cost, and the server-side web searches the call made). A research run's result and its watchlist_research_periodic_ran audit row carry a usage block summing its calls per source, and GET /audit/usage (and the usage page) gain a by-source breakdown.

Removed

  • The xcrawl scrape service and everything that depended on it. The research run no longer has a post-dedup verify pass or an optional read-before-cite scrape loop, and the watch policy no longer scores a "verified" point; the XCRAWL_*, EXTERNAL_RESEARCH_VERIFY_*, RESEARCH_AGENTIC_*, RESEARCH_SCRAPE_* and RESEARCH_LOOP_* settings are gone (a deployment that still sets them is unaffected). Every watchlist fetch is the keyless, SSRF-guarded httpx fetcher: a page watch written under the old fetch: xcrawl marker re-captures its baseline on the next poll without reporting a change, and an rss target that is not a feed becomes a page_watch when the page already fetched by the feed check has readable text (otherwise it is rejected, as before).

Changed

  • The research routing pass no longer carries the watchlist write tools (add_watchlist_entry, tune_watchlist_entry, remove_watchlist_entry): every watch a research run creates goes through the watchlist policy. tune_watchlist_entry refuses mode / enabled changes on a pending suggestion, URL targets are SSRF-checked at insert time for every kind, and deleting a watch now removes its signal history in the same transaction (the foreign key previously made any polled watch undeletable).
  • Watchlist rows carry an origin (manual | executive | research | research_proposed); the research skip-if-unchanged fingerprint hashes only enabled, active rows so approving or declining a suggestion never triggers a fresh 7-specialist run; the research turn shows each watch's origin, fired/dismissed counts and trust, plus declined targets.
  • Daily briefs are bounded to "since the last delivered brief": what's new, what the Executive handled overnight, who we're waiting on, one top call, and a one-line carried-over count. An unchanged day sends "Nothing new since yesterday's brief — N items still waiting on you." and skips the model call (PRINCIPAL_BRIEF_SUPPRESS_UNCHANGED; force_full on a manual run bypasses it). The EoD digest now excludes monitoring noise.
  • The daily reflection sees yesterday's standup and is told not to repeat it; the watchlist research council's forced re-run moves from daily to weekly (WATCHLIST_RESEARCH_MAX_STALENESS_HOURS=168); stalled-workflow nudges stop after NUDGE_MAX_PER_SCOPE=3 per item.
  • Alerts created by the Executive's create_alert tool now land with their routed_to_person_id and department:<slug> tag (previously every triage-born alert fell to the principal as "unrouted").

Upgrading

  • All schema changes are additive columns with defaults (alerts table; watchlist.origin, plus two new tables watchlist_declines and watchlist_policy_outcomes); rolling the image back is safe. No config changes are required.
  • The research council now adds a watch on its own only when it is grounded in your company data. Name your vendors and tickers on the Company Profile page ("External Dependencies") so status pages and filings for them land without asking; everything else arrives as a suggestion on /watchlist. Existing watches are untouched (origin=manual) and never auto-disabled.
  • After upgrading, the first scheduler sweep expires alerts past their TTL (audited as alert_sweep, reversible with reopen); raise ALERT_TTL_DAYS_* (or set 0) beforehand to keep an old backlog. The review job starts acting within ALERT_REVIEW_MIN_AGE_HOURS (2 h) — set ALERT_REVIEW_ENABLED=false to turn it off, or ALERT_REVIEW_MAX_MOVES_PER_SCAN=0 to keep it annotate-only.
  • Briefs get shorter and send a one-liner on unchanged days (PRINCIPAL_BRIEF_SUPPRESS_UNCHANGED=false restores daily full briefs).
  • docker/docker-compose.yml now binds the API to 127.0.0.1:8000 instead of every host interface. The UI reaches the API over the compose network, so nothing in the stack needed the public port. If you were calling :8000 directly from another host, put a reverse proxy in front of it and set BACKEND_SHARED_SECRET and OE_PUBLIC_DEPLOYMENT=1.

0.1.0 - 2026-06-30

Initial public release.

Added

  • Multi-agent "Executive" system: a single coherent executive persona backed by eight specialist sub-agents, powered by the Anthropic Claude API.
  • Python backend (packages/core) — FastAPI service, orchestrator, specialist agents, ChromaDB-backed RAG, CLI, and prompt-caching layer.
  • Next.js 15 web UI (packages/ui), including the static /architecture page.
  • Curated MBA knowledge base (knowledge/) and eval suite (evals/).
  • Optional integrations: Slack, Discord, and email.
  • Docker deployment configuration.
  • Open-source project setup: Apache-2.0 license, contribution guide, code of conduct, security policy, issue/PR templates, and CI.