[kovtcharov-amd]

GAIA v0.24.0 Release Notes

Breaking Changes

• Thirteen per-task agents were removed — analyst, blender, browser, code, doc-search, docker, docqa, emr, fileio, jira, routing, sd, summarize; it's a SKILL.md in hub/skills/ on the flagship now. Installing one by id fails — use gaia hub install gaia --trust and load the matching skill. (PR #2995)
• The TUI control API dropped its hub-browser fields — view is now splash | preflight | chat; hub_tab, hub_tab_index, selected_agent_id, visible_agent_ids, filtering, can_return_to_hub are gone, replaced by blocker. Anyone driving the TUI over this API needs to update. (PR #3120)
• @tool(name=..., description=...) no longer works — the decorator rejects unrecognized keywords instead of silently discarding them. Use @tool(); name and description come from the function. (PR #3252)
• gaia telegram start now requires --allowed-users, no opt-out — it used to start (and reply to anyone) without one. (PR #3306)
• gaia api --host beyond loopback with no key now refuses to start, and GAIA_API_KEY gates chat completions too. (PR #3612)

What's New

• gaia hub install gaia --trust actually installs and runs the flagship agent — the daemon had no sidecar entry for it, the installers didn't ship its binary, and gaia init set up the wrong agent. All three are fixed. (PRs #3031, #3146, #3330, #3125, #3120)

• The agent draws pictures, summarizes meetings, and orients itself first — "draw me a red bicycle" reaches the image tools; summarize this meeting: standup.mp4 returns a transcript with action items; and every task scans the project first. Skills load themselves, and corrections are learned into memory, not hand-edited into SKILL.md. (PRs #3073, #3597, #3405, #3010, #3023)

• Every answer states whether it was checked — the agent names what it ran and what passed, not just reporting success. --trace in the TUI records every tool call so a wrong answer can be replayed. (PRs #3397, #3579)

• The TUI takes the mouse, draws to full width, and can go cloud — wheel-scroll, clickable links, no more 80-column wrapping, and a local, Fireworks, or AMD Gateway model via Lemonade. (PRs #3699, #3006, #3672)

• Run Lemonade without the machine-wide installer — gaia lemonade embedded start runs your own Lemonade Server, no admin rights needed. (PR #3121)

• Mine your own Claude Code sessions for usage stats — a deterministic, LLM-free pipeline turns your ~/.claude/projects/ transcripts into tool-usage, failure, and cost reports. See the Session Analytics guide. (PRs #3111, #3782)

Security

• Any web page you visited could drive GAIA's local servers — a cross-origin request could start a tunnel, poison your RAG store, or send mail; a document link could make the chat agent fetch cloud metadata or the daemon's loopback API. Middleware now enforces Origin/Host checks everywhere, and web tools refuse private/loopback redirects. (PRs #3366, #3612)
• A Telegram bot with no allowlist replied to anyone on the platform (the documented default) — off-allowlist users are now refused before a session or model call; the bot's tool surface is chat-only. (PR #3306)
• Attaching a document could silently widen the agent's read scope to its whole parent directory — the grant is now the file itself; reads gained a credential denylist, and writes refuse shell rc files, PowerShell profiles, and git hooks. (PR #3594)
• A chat turn could plant system/permission rows into every future prompt — only onboarding/admin paths can write them now. (PR #3614)

Bug Fixes

A selection of the user-visible fixes this release — the full list is in the changelog below.

• The flagship agent's core model floor now matches what can actually run it (PR #3055) — installs no longer point at a Lemonade version too old to serve the agent.
• A one-shot /query run no longer leaks an agent process (PR #3079).
• The npm sidecar no longer attaches to a foreign server (PR #3078) — it now verifies the server it finds is actually its own before using it.
• A turn parked on a confirmation prompt no longer hangs when stdin closes (PR #3080) — the agent ends the turn instead of waiting forever.
• Telegram's background polling survives (PR #3272) — the adapter no longer silently stops receiving messages.
• gaia uninstall --purge removes the embedded Lemonade runtime it installed (PR #3277) — previously a full purge left ~/.gaia/lemonade behind entirely; a live embedded daemon must be stopped first.
• Outlook and Gmail search operators translate correctly for the email agent (PRs #3042, #3562, #3619) — a query that used to fail silently or match the wrong messages against Outlook's Graph API now works or fails loudly.
• RAG indexing honors a document's actual success flag (PR #3678) — a failed index no longer gets reported as indexed at every call site that checked it.
• Claude tool-call history survives across turns (PR #3410) — a multi-turn tool conversation no longer loses earlier tool results.
• An ambiguous edit is refused instead of silently changing the first match (PR #3396) — edit_file asks for a more specific match rather than guessing which occurrence you meant.
• A port collision between the MCP bridge, Agent UI, TUI, and Telegram is resolved (PR #3604) — the four no longer fight over the same default port.
• Telegram reports unsupported media types instead of failing silently (PR #3263).
• GET /health reports real component status (PR #2601) — it used to report ok even with a dead Lemonade backend; it now reports LLM and RAG status too.
• A .bak backup file could be imported as the module it backed up (PR #3748) — .bak now goes last in the filename, so pytest no longer tries to collect a test file's own backup.
• A PowerShell sink could execute untrusted text instead of just displaying it (PR #3359) — an apostrophe in an indexed document name, or in a notify_desktop message, ran as code; both now pass values out-of-band.
• --scopes could request an OAuth scope outside a connector's declared ceiling (PR #3247) — the CLI's verbatim-scopes path skipped the check the Agent UI already had; both now enforce it.

Contributors

Thanks to the community members who contributed to this release:

• @mikemikimike — Telegram background-polling and media-handling fixes, embedded-Lemonade uninstall cleanup, and TUI init-process reaping (PRs #3272, #3277, #3271)
• @itomek — custom-agent guide fixes and CI reliability work, including provisioning the dev install in Claude Code cloud sessions (PRs #3317, #3033)
• @Hashim1999164 — hub agent pages showing the GAIA install path alongside npm, structured JSON for unexpected email API errors, and Outlook Graph query-escaping fixes (PRs #3151, #3046, #3021)
• @AmirF194 — Outlook/Gmail search-operator translation for the email agent (PRs #3042, #3562, #3619)
• @sujikathir — the /health component-status endpoint and a Claude tool-history preservation fix (PRs #2601, #3410)
• @SIBTAIN-ASAD — Telegram unsupported-media-type reporting and test coverage (PR #3263)
• @wangzhengzhuo05 — MCP/Telegram/TUI port-collision fix across the bridge, Agent UI, and daemon (PR #3604)
• @amd-pworfolk — stopped a non-LLM model's ctx_size from leaking into get_status() (PR #3595)

Full Changelog

230 commits since v0.23.0:

Expand full changelog (230 commits)

• 40e0dd89d — docs(reference): add a user guide for Claude Code session analytics (docs(reference): add a user guide for Claude Code session analytics #3782)
• f90fa5aae — fix(docs): stop misclassifying GitHub anti-abuse resets as broken links (fix(docs): stop misclassifying GitHub anti-abuse resets as broken links #3800)
• 87313af75 — fix(daemon): reject a repo-root --dev-src-dir with a corrected path (fix(daemon): reject a repo-root --dev-src-dir with a corrected path #3796)
• cad3363f7 — fix(agents): isolate log records from abandoned tool-timeout workers (fix(agents): isolate log records from abandoned tool-timeout workers #3795)
• 43dbce921 — fix(connectors): forward_connection uses the shared scope resolver (fix(connectors): forward_connection uses the shared scope resolver #3792)
• 0c301bae3 — fix(tui): abandon queued follow-up on second Esc, add stall nudge (fix(tui): abandon queued follow-up on second Esc, add stall nudge #3798)
• 99caeaaf7 — fix(email-agent): close draft/RSVP/event gap in fabricated-success guard (fix(email-agent): close draft/RSVP/event gap in fabricated-success guard #3797)
• 8eb53c6b7 — fix(tui): name the buckets behind email pre-scan's +N more (fix(tui): name the buckets behind email pre-scan's +N more #3794)
• 5597e6a57 — fix(connectors): turn on include_granted_scopes for Google OAuth (fix(connectors): turn on include_granted_scopes for Google OAuth #3786)
• 76a7cde29 — fix(webui): show agent version in Hub Details modal (fix(webui): show agent version in Hub Details modal #3784)
• 1474b12e9 — fix(tests): assert no render card for pre_scan_inbox (fix(tests): assert no render card for pre_scan_inbox #3785)
• 70de05022 — fix(email-agent): assert SCHEMA_VERSION, not a hardcoded literal (fix(email-agent): assert SCHEMA_VERSION, not a hardcoded literal #3787)
• ddfdcabea — fix(tests): skip live release-asset test on transient network drops (fix(tests): skip live release-asset test on transient network drops #3788)
• 4740ec97f — fix(email): point README at gaia init instead of removed CLI (fix(email): point README at gaia init instead of removed CLI #3783)
• ff62b4d54 — fix(agent): strip whitespace from tool names before dispatch (fix(agent): strip whitespace from tool names before dispatch #3744)
• f39ead7ad — fix(agent_loop): stop spending hourly budget on idle no-op ticks (fix(agent_loop): stop spending hourly budget on idle no-op ticks #3745)
• 1a93f3e20 — fix: apply reviewed maintenance repairs for fix(ci): run the full suite on stacked PRs, not just PRs into main #3372 (fix: apply reviewed maintenance repairs for #3372 #3713)
• fe863b5d9 — fix: apply reviewed maintenance repairs for fix(eval): make the agent eval gate's report mode actually report #3613 (fix: apply reviewed maintenance repairs for #3613 #3717)
• fa02dc063 — fix: apply reviewed maintenance repairs for fix(eval): capture stdout when a scenario subprocess fails #3375 (fix: apply reviewed maintenance repairs for #3375 #3714)
• 3c32386ce — fix(security): put .bak last so a backup stops looking like a Python module (fix(security): put .bak last so a backup stops looking like a Python module #3748)
• c11e2bf85 — test(installer): expect the not-executed flag on a refused MCP tool call (test(installer): expect the not-executed flag on a refused MCP tool call #3736)
• 2cbde094b — fix(vlm): raise when vision extraction fails instead of indexing the error (fix(vlm): raise when vision extraction fails instead of indexing the error #3722)
• 8c4661c38 — fix(agent,memory): let a bare recall() browse instead of erroring (fix(agent,memory): let a bare recall() browse instead of erroring #3705)
• 7bd576723 — fix(agent): stop the verification footer claiming a rejected command ran (fix(agent): stop the verification footer claiming a rejected command ran #3703)
• 5bb9e53f0 — fix(tools): keep a finished edit reported as finished when the diff won't render (fix(tools): keep a finished edit reported as finished when the diff won't render #3702)
• 044a741b3 — fix(tools): search the user's project, not the directory the sidecar was spawned in (fix(tools): search the user's project, not the directory the sidecar was spawned in #3711)
• 9fc740074 — fix(tools): describe the edit and write tools by what agents actually edit (fix(tools): describe the edit and write tools by what agents actually edit #3712)
• eefd33dde — fix(rag): stop indexing a damaged PDF page as a blank one (fix(rag): stop indexing a damaged PDF page as a blank one #3723)
• 3d0dfbee1 — fix(ui): let a document on a reconnected drive come back (fix(ui): let a document on a reconnected drive come back #3727)
• cd3dbe6de — fix(talk): say when the microphone fails instead of printing "Listening…" (fix(talk): say when the microphone fails instead of printing "Listening…" #3724)
• 7c733bb01 — fix(hub): stop recording a failed first install as installed (fix(hub): stop recording a failed first install as installed #3726)
• 140619559 — feat(tui): scroll with the wheel, click links, and print only measured ttft/tok-s (feat(tui): scroll with the wheel, click links, and print only measured ttft/tok-s #3699)
• 8113fd043 — fix(code-index): let a session index more than one repository (fix(code-index): let a session index more than one repository #3728)
• d311ac39e — fix(tools): describe tool arguments at the slot the model fills (fix(tools): describe tool arguments at the slot the model fills #3583)
• 62e52f8e5 — fix(ui): let the operator declare document roots beyond $HOME (fix(ui): let the operator declare document roots beyond $HOME #3586)
• c57db1f0d — fix(tools): honor find_files scope on the index path and stop doubling "error:" (fix(tools): honor find_files scope on the index path and stop doubling "error:" #3671)
• cb77a334f — fix(agents): retry instead of answering with raw tool-call JSON (fix(agents): retry instead of answering with raw tool-call JSON #3598)
• 41cd64ec2 — fix(llm): keep tool calls on a truncated assistant turn during overflow retry (fix(llm): keep tool calls on a truncated assistant turn during overflow retry #3582)
• 36b8aa55d — feat(skills): make image generation reachable from the flagship (feat(skills): make image generation reachable from the flagship #3073)
• 600831f9e — ci(vscode): drop the base-branch filter that now fails the lint gate (ci(vscode): drop the base-branch filter that now fails the lint gate #3719)
• 1c23713bd — fix(ci): run the full suite on stacked PRs, not just PRs into main (fix(ci): run the full suite on stacked PRs, not just PRs into main #3372)
• b8e0496b1 — fix(eval): capture stdout when a scenario subprocess fails (fix(eval): capture stdout when a scenario subprocess fails #3375)
• d5d4a815c — fix(tui,webui): stop the real agent on cancel; keep answer content (fix(tui,webui): stop the real agent on cancel; keep answer content #3627)
• 24c0ee4fd — fix(agent-gaia): name a core floor that can actually run the agent (fix(agent-gaia): name a core floor that can actually run the agent #3055)
• eee4666cd — fix(security): make the file boundary the file, and hold it on reads (fix(security): make the file boundary the file, and hold it on reads #3594)
• 46a41f9ec — fix(chat): restore recent context in long sessions (fix(chat): restore recent context in long sessions #3651)
• 0d82616a9 — fix(tui): reject unsupported daemon permission bypass (fix(tui): reject unsupported daemon permission bypass #3654)
• c3284b464 — fix(npm): resolve Windows shims before launching the TUI (fix(npm): resolve Windows shims before launching the TUI #3653)
• a07b87e40 — feat(skills): let a gh write ask instead of fail (feat(skills): let a gh write ask instead of fail #3009)
• 1070ed226 — fix(api,mcp,web): refuse foreign origins and internal fetch targets (fix(api,mcp,web): refuse foreign origins and internal fetch targets #3612)
• e373a19df — chore(lemonade): remove the dead lemonade-server serve CLI from the last surfaces (chore(lemonade): remove the dead lemonade-server serve CLI from the last surfaces #3617)
• 305ae88e9 — fix(rag): preserve indexed documents when persistence fails (fix(rag): preserve indexed documents when persistence fails #3652)
• b76647e72 — fix(mcp): retrieve complete Agent UI transcripts (fix(mcp): retrieve complete Agent UI transcripts #3667)
• b7884b46d — fix(rag): honor index_document's success flag at every truthy call site (fix(rag): honor index_document's success flag at every truthy call site #3678)
• 1d3b2b415 — feat(tools): tell the agent its edit tools exist (feat(tools): tell the agent its edit tools exist #3602)
• 86071110a — fix(eval): make the agent eval gate's report mode actually report (fix(eval): make the agent eval gate's report mode actually report #3613)
• 404107919 — fix(memory): refuse chat-minted privileged rows and distil before pruning (fix(memory): refuse chat-minted privileged rows and distil before pruning #3614)
• 2fc19e5ee — feat(tui): choose local, Fireworks, or AMD Gateway via Lemonade (feat(tui): choose local, Fireworks, or AMD Gateway via Lemonade #3672)
• a5e16d0b9 — test(lemonade): assert four external contracts the mocks were faking (test(lemonade): assert four external contracts the mocks were faking #3615)
• 947d05225 — fix(schedule): honor day constraints in recurring timers (fix(schedule): honor day constraints in recurring timers #3655)
• 5155591ad — fix(chat): retain admission slots for detached runs (fix(chat): retain admission slots for detached runs #3657)
• 13f61aebc — fix(broker): wake queued requests at lease expiry (fix(broker): wake queued requests at lease expiry #3660)
• cb8c847a7 — fix(vscode): preserve split streaming tool calls (fix(vscode): preserve split streaming tool calls #3661)
• 65499dabf — fix(chat): load complete transcripts across message pages (fix(chat): load complete transcripts across message pages #3664)
• d84f98474 — ci(eval): give the persistent Lemonade task the same launch env as every other start path (ci(eval): give the persistent Lemonade task the same launch env as every other start path #3577)
• 42377da7a — fix(llm): don't let a non-LLM model's ctx_size leak into get_status() (fix(llm): don't let a non-LLM model's ctx_size leak into get_status() #3595)
• e6a9efcf5 — Error when a named find_files folder is missing (Error when a named find_files folder is missing #3580)
• 4d47ae9a9 — feat(agent): transcribe and summarize meeting recordings locally (feat(agent): transcribe and summarize meeting recordings locally #3597)
• cfc3d2959 — fix(email): don't treat a colon word inside a quoted phrase as an Outlook operator (fix(email): don't treat a colon word inside a quoted phrase as an Outlook operator #3619)
• e8f5bc898 — fix(mcp,telegram): resolve port 8765 collision across bridge, agent-ui, tui and telegram health (fix(mcp,telegram): resolve port 8765 collision across bridge, agent-ui, tui and telegram health #3604)
• a2df094e3 — fix(llm): preserve Claude tool history (fix(llm): preserve Claude tool history #3410)
• d9e381b9d — fix(tui): reap cancelled init processes (fix(tui): reap cancelled init processes #3271)
• d7cadd3e0 — fix(telegram): keep background polling alive (fix(telegram): keep background polling alive #3272)
• b81f9e1b0 — fix: uninstall embedded Lemonade runtime (fix: uninstall embedded Lemonade runtime #3277)
• 146dd63cc — fix(tui): send caller checkout to daemon ensure (fix(tui): send caller checkout to daemon ensure #3268)
• 4ce7c2662 — fix(email): allow move_to_label to restore inbox (fix(email): allow move_to_label to restore inbox #3267)
• 36cab66de — fix(connectors): preserve grants in device login (fix(connectors): preserve grants in device login #3266)
• 176faf8eb — fix(email): report calendar listing truncation (fix(email): report calendar listing truncation #3264)
• ca2a5086d — fix(email): honor sender preference ordering (fix(email): honor sender preference ordering #3257)
• 659cb0cde — fix(email): surface briefing cache age (fix(email): surface briefing cache age #3256)
• 7304f5e86 — ci(release): let the backend test step fail the release (ci(release): let the backend test step fail the release #3573)
• 07a3749c0 — fix(ci): report runner outages in minutes, not as a silent cancel a day later (fix(ci): report runner outages in minutes, not as a silent cancel a day later #3411)
• 68f10a8fa — fix(agents): reject unknown @tool keyword arguments (fix(agents): reject unknown @tool keyword arguments #3252)
• c45593aed — fix(email-tests): prevent headless keyring hangs (fix(email-tests): prevent headless keyring hangs #3249)
• 543b11ec6 — fix(skills): make skill tools work on the Claude provider (fix(skills): make skill tools work on the Claude provider #3143)
• f83e96c14 — feat(agent): materialize a project map at task start (feat(agent): materialize a project map at task start #3405)
• 29e5d977c — feat(tui): add --trace to record what the agent actually did (feat(tui): add --trace to record what the agent actually did #3579)
• 38bc3669b — docs(troubleshooting): explain the audit-module antivirus false positive (docs(troubleshooting): explain the audit-module antivirus false positive #3348)
• f3d739485 — fix(email): fail loudly on Outlook operators Graph cannot translate (fix(email): fail loudly on Outlook operators Graph cannot translate #3562)
• af7d492ed — fix(eval): drive the agent eval with the OAuth token, and probe it for real (fix(eval): drive the agent eval with the OAuth token, and probe it for real #3403)
• 0c23e13a9 — fix(tools): refuse an ambiguous edit instead of changing the first match (fix(tools): refuse an ambiguous edit instead of changing the first match #3396)
• 4ccaa0738 — feat(agent): state whether an answer was verified, on every exit path (feat(agent): state whether an answer was verified, on every exit path #3397)
• 4c229d434 — fix(api,web): repair gaia api chat completions, HTTPS SNI, and download overwrite (fix(api,web): repair gaia api chat completions, HTTPS SNI, and download overwrite #3364)
• 456aef994 — chore(deps): bump anthropics/claude-code-action from 1.0.210 to 1.0.216 in the github-actions group across 1 directory (chore(deps): bump anthropics/claude-code-action from 1.0.210 to 1.0.216 in the github-actions group across 1 directory #3409)
• a87c3f631 — chore(deps-dev): bump the agent-ui-dependencies group in /src/gaia/apps/webui with 5 updates (chore(deps-dev): bump the agent-ui-dependencies group in /src/gaia/apps/webui with 5 updates #3407)
• 7e7a2684d — docs(plans): measure GAIA's agent-loop gaps against 345 real sessions (docs(plans): measure GAIA's agent-loop gaps against 345 real sessions #3391)
• d09a7a64e — feat(skills): learn a skill's corrections into memory, never into the file (feat(skills): learn a skill's corrections into memory, never into the file #3023)
• c53918b75 — docs(claude): correct stale commands, the GaiaAgent status, and the CLI list (docs(claude): correct stale commands, the GaiaAgent status, and the CLI list #3311)
• e274986a1 — fix(skills,tools): stop two tools deleting data while reporting success (fix(skills,tools): stop two tools deleting data while reporting success #3358)
• 74916c5f5 — fix(cli): stop uninstall and kill --port destroying what they do not own (fix(cli): stop uninstall and kill --port destroying what they do not own #3361)
• 0ec6ee13e — fix(connectors): make the per-agent grant check real and bounded (fix(connectors): make the per-agent grant check real and bounded #3363)
• 9bd6bbadd — chore(deps-dev): bump the python-dependencies group with 2 updates (chore(deps-dev): bump the python-dependencies group with 2 updates #3408)
• a65bc4556 — chore(deps-dev): bump electron from 44.0.0 to 44.1.1 in the root-npm-dependencies group (chore(deps-dev): bump electron from 44.0.0 to 44.1.1 in the root-npm-dependencies group #3406)
• b4f2c096f — fix(telegram): scope the bot's tool surface and stop the allowlist failing open (fix(telegram): scope the bot's tool surface and stop the allowlist failing open #3306)
• d7470a8fc — fix(ci): stop the Claude audits re-filing the same finding every run (fix(ci): stop the Claude audits re-filing the same finding every run #3310)
• dd98c0ff4 — fix(security): stop PowerShell sinks from executing untrusted text (fix(security): stop PowerShell sinks from executing untrusted text #3359)
• 833ca3d06 — fix(ui): stop any web page from driving the Agent UI backend (fix(ui): stop any web page from driving the Agent UI backend #3366)
• e9f28b311 — fix(release): make the v0.23.1 tag pass its own publish gates (fix(release): make the v0.23.1 tag pass its own publish gates #3354)
• abe87edc3 — Devlab/dispatch passing (Devlab/dispatch passing #3344)
• 5b284b4da — feat(init): make gaia init the one setup path for the flagship (feat(init): make gaia init the one setup path for the flagship #3330)
• 1ae7dad84 — fix(connectors): enforce catalog scope ceiling for authorization (fix(connectors): enforce catalog scope ceiling for authorization #3247)
• 0525d5b0a — fix(chat): surface unavailable memory at startup (fix(chat): surface unavailable memory at startup #3153)
• 695e92c58 — docs(custom-agent): make the guide's agent examples actually run (docs(custom-agent): make the guide's agent examples actually run #3317)
• a75904de1 — chore(deps-dev): bump the agent-ui-dependencies group in /src/gaia/apps/webui with 2 updates (chore(deps-dev): bump the agent-ui-dependencies group in /src/gaia/apps/webui with 2 updates #3314)
• a6a25f29c — chore(deps): bump zod from 4.4.3 to 4.5.4 in the root-npm-dependencies group (chore(deps): bump zod from 4.4.3 to 4.5.4 in the root-npm-dependencies group #3313)
• ed32a2d4d — fix(daemon): treat an unreaped sidecar as stopped, not as a survivor (fix(daemon): treat an unreaped sidecar as stopped, not as a survivor #3229)
• 211f08c5b — Use the workspace root lockfile for the example Electron app (Use the workspace root lockfile for the example Electron app #3301)
• 1886dd363 — test(conftest): stop the unit suite opening the developer's real browser (test(conftest): stop the unit suite opening the developer's real browser #3305)
• 72241b2f2 — Release v0.23.1 (Release v0.23.1 #3054)
• 9b26827ca — fix: keep source-less agents compatible with skill discovery (fix: keep source-less agents compatible with skill discovery #3244)
• 0fb49f1d6 — fix: preserve forwards across transient mint failures (fix: preserve forwards across transient mint failures #3245)
• 1dec2462f — fix(connectors): record only granted agent scopes (fix(connectors): record only granted agent scopes #3248)
• 3e90f15d5 — fix(agent): use device profile for context overflow (fix(agent): use device profile for context overflow #3253)
• 18a8bcb89 — fix: guide daemon version skew upgrades (fix: guide daemon version skew upgrades #3246)
• 39af8ef9f — fix(ui): report Lemonade probe failures accurately (fix(ui): report Lemonade probe failures accurately #3250)
• 0db152ebc — fix(daemon): honor extended agent query timeouts (fix(daemon): honor extended agent query timeouts #3251)
• 2dedc7654 — fix(tui): preserve partial-success batch outcomes (fix(tui): preserve partial-success batch outcomes #3255)
• c5b00e85b — fix(telegram): report unsupported media types (fix(telegram): report unsupported media types #3263)
• bc613dc85 — fix(hub): stop calling an app's npm package a frozen agent sidecar (fix(hub): stop calling an app's npm package a frozen agent sidecar #3299)
• 821667c76 — fix(skills): normalize common plural stems (fix(skills): normalize common plural stems #3270)
• 8ad4ee469 — fix(email): parse grouped Outlook duration queries (fix(email): parse grouped Outlook duration queries #3269)
• 30b8246ad — test: cover registered Telegram handlers (test: cover registered Telegram handlers #3273)
• 80cf9eb00 — test: verify embedded backend install contract (test: verify embedded backend install contract #3274)
• 9b40c3522 — ci: run embedded Lemonade checksum guard (ci: run embedded Lemonade checksum guard #3275)
• 5b45a7c11 — fix: keep skill discovery loader available (fix: keep skill discovery loader available #3276)
• 1ff94ae6c — fix: report RAG indexing failures accurately (fix: report RAG indexing failures accurately #3278)
• 0c4d725a0 — fix: report missing API server on stop (fix: report missing API server on stop #3279)
• 811e0bc04 — fix(talk): honor restart voice command (fix(talk): honor restart voice command #3280)
• 869f0bf41 — fix: fail Telegram startup when dependency is missing (fix: fail Telegram startup when dependency is missing #3283)
• 31fcdd03c — fix(mcp): expose TUI state wait matcher (fix(mcp): expose TUI state wait matcher #3284)
• f2dfcd8cb — fix(hub): stop offering a GAIA install a non-agent card rules out (fix(hub): stop offering a GAIA install a non-agent card rules out #3232)
• 6ffee9e02 — chore(deps-dev): bump the agent-ui-dependencies group in /src/gaia/apps/webui with 5 updates (chore(deps-dev): bump the agent-ui-dependencies group in /src/gaia/apps/webui with 5 updates #3156)
• 671000d8e — chore(deps): bump the root-npm-dependencies group with 3 updates (chore(deps): bump the root-npm-dependencies group with 3 updates #3155)
• 438733605 — chore(deps): bump anthropics/claude-code-action from 1.0.206 to 1.0.210 in the github-actions group (chore(deps): bump anthropics/claude-code-action from 1.0.206 to 1.0.210 in the github-actions group #3157)
• 0c28442a9 — fix(tui): preserve seeded transport for wheel agents (fix(tui): preserve seeded transport for wheel agents #3285)
• 41a00915b — fix(tui): gate Claude launches on credentials (fix(tui): gate Claude launches on credentials #3286)
• 25dd767da — fix(ci): the eval gate skipped every category when the embedder failed (fix(ci): the eval gate skipped every category when the embedder failed #3158)
• fdcdbff4c — fix(website): Mac visitors on Safari got no install button (fix(website): Mac visitors on Safari got no install button #3144)
• 95f874817 — perf(gaia): send the flagship only the tools a turn needs (perf(gaia): send the flagship only the tools a turn needs #3008)
• eb764bf5a — feat(installer): the one-line install now brings the agent with it (feat(installer): the one-line install now brings the agent with it #3146)
• ea0e5d924 — feat(cpp): skill sets and gaia-agent.yaml wiring (feat(cpp): skill sets and gaia-agent.yaml wiring #2931)
• 5b37cb138 — Route the Strix Halo jobs to the Ryzen Dev Lab pool (Route the Strix Halo jobs to the Ryzen Dev Lab pool #3262)
• 0047af97d — ci(claude): skip Claude PR review for Dependabot PRs (ci(claude): skip Claude PR review for Dependabot PRs #3259)
• 11039b64e — test(daemon): stop the hub stray-process test orphaning a sleep child (test(daemon): stop the hub stray-process test orphaning a sleep child #3230)
• a338b94f4 — test(telegram): cover media ingest paths (test(telegram): cover media ingest paths #3150)
• a7bcf5e11 — Show GAIA install alongside npm on hub agent pages (Show GAIA install alongside npm on hub agent pages #3151)
• ecae8f183 — test(telegram): pin the _allowed truth table including the fail-open (test(telegram): pin the _allowed truth table including the fail-open #3051)
• 8e57b5197 — docs(readme): note the Email agent's CLI install path (docs(readme): note the Email agent's CLI install path #3041)
• b9330deba — fix(email): translate Gmail-style operators for Outlook Graph search (fix(email): translate Gmail-style operators for Outlook Graph search #3042)
• 3b23ac2b3 — feat(skills): load the right skill without the user knowing its name (feat(skills): load the right skill without the user knowing its name #3010)
• d4666b8bd — feat(lemonade): run a private, self-contained Lemonade Server (feat(lemonade): run a private, self-contained Lemonade Server #3121)
• 793fb4d42 — perf(ci): cut the C++ dependency cache from 407 MB to ~109 MB (perf(ci): cut the C++ dependency cache from 407 MB to ~109 MB #3142)
• 2cd894579 — fix(telegram): apply the allowlist to every handler, not just messages (fix(telegram): apply the allowlist to every handler, not just messages #3148)
• c128b18f6 — fix(cpp): unblock the C++ gate — split the third-party skill corpus from the conformance one (fix(cpp): unblock the C++ gate — split the third-party skill corpus from the conformance one #3129)
• 55c977006 — feat(factory): harvest Claude Code sessions into usage statistics (feat(factory): harvest Claude Code sessions into usage statistics #3111)
• 6db50643d — chore(deps): bump the github-actions group across 1 directory with 3 updates (chore(deps): bump the github-actions group across 1 directory with 3 updates #3049)
• c7989b953 — fix(mcp): port servers to MCP SDK 2 (fix(mcp): port servers to MCP SDK 2 #3047)
• ee61b74fe — Return structured JSON for unexpected email API errors (Return structured JSON for unexpected email API errors #3046)
• f2d8291c3 — feat(installer): double-click installers for the terminal hub (feat(installer): double-click installers for the terminal hub #3125)
• f601621d1 — fix(agent-gaia): stop one-shot /query runs leaking an agent each (fix(agent-gaia): stop one-shot /query runs leaking an agent each #3079)
• 4a6561c86 — chore(lemonade): pin Lemonade 11.8.1 (chore(lemonade): pin Lemonade 11.8.1 #3117)
• c51a1c520 — test(telegram): cover access and session isolation (test(telegram): cover access and session isolation #3052)
• 98043fb56 — feat(tui): boot straight into the flagship agent, retire the hub browser (feat(tui): boot straight into the flagship agent, retire the hub browser #3120)
• 5bb64a6ba — chore(deps-dev): bump the agent-ui-dependencies group across 1 directory with 8 updates (chore(deps-dev): bump the agent-ui-dependencies group across 1 directory with 8 updates #3048)
• 51c298e1a — fix(tui): stop --use-claude from starting Lemonade, and name the Claude model (fix(tui): stop --use-claude from starting Lemonade, and name the Claude model #3005)
• 38423c84c — fix(ci): unblock the walkthrough's wheel install, rein in the authz lens's subagents (fix(ci): unblock the walkthrough's wheel install, rein in the authz lens's subagents #3109)
• abb8ddea8 — ci(release): require GAIA_HUB_PUBLISH_URL for hub publish, drop dead fallback (ci(release): require GAIA_HUB_PUBLISH_URL for hub publish, drop dead fallback #3032)
• 88299ea49 — feat(website): lead with the flagship agent, and say what it actually does (feat(website): lead with the flagship agent, and say what it actually does #3093)
• eda11b778 — fix(llm): suggest gaia init on macOS (fix(llm): suggest gaia init on macOS #3027)
• 11f4da758 — Escape quotes inside Outlook Graph search queries (Escape quotes inside Outlook Graph search queries #3021)
• 1a3d0f0bd — fix(packaging): add realtime dependencies to talk extra (fix(packaging): add realtime dependencies to talk extra #2954)
• ae0b64e9f — chore(deps-dev): bump electron from 43.2.0 to 43.4.1 in the root-npm-dependencies group across 1 directory (chore(deps-dev): bump electron from 43.2.0 to 43.4.1 in the root-npm-dependencies group across 1 directory #2871)
• b95c66dff — test(memory): assert the reconcile 503 branch in every environment (test(memory): assert the reconcile 503 branch in every environment #2868)
• 708f806a0 — docs(playbooks): simplify index page (docs(playbooks): simplify index page #2957)
• 4ac8f1075 — fix(cli): drop non-constructor CLI flags before creating GaiaCliClient (fix(cli): drop non-constructor CLI flags before creating GaiaCliClient #2976)
• 191a317f2 — feat(api): report component health (feat(api): report component health #2601)
• 40ed1513b — chore(deps-dev): bump electron from 43.2.0 to 43.4.1 in /src/gaia/apps/example/webui in the example-app-dependencies group across 1 directory (chore(deps-dev): bump electron from 43.2.0 to 43.4.1 in /src/gaia/apps/example/webui in the example-app-dependencies group across 1 directory #2869)
• dac81b60c — docs(tui): tell people how to install the terminal hub (docs(tui): tell people how to install the terminal hub #3039)
• 643f78432 — feat(tui): lay out to the real terminal width (feat(tui): lay out to the real terminal width #3006)
• ba64ed1e3 — fix(installer): install the extras the daemon needs to start (fix(installer): install the extras the daemon needs to start #3056)
• 25c137275 — fix(ci): make the Claude audits produce output, not just run green (fix(ci): make the Claude audits produce output, not just run green #3063)
• ee64d5cd2 — docs(agent-gaia): correct the npm sidecar docs against the code (docs(agent-gaia): correct the npm sidecar docs against the code #3076)
• 625478393 — fix(tui): don't run an unverified binary from the hub install root (fix(tui): don't run an unverified binary from the hub install root #3075)
• a6c68c8a4 — fix(agent-email): fail closed when a request omits its Host header (fix(agent-email): fail closed when a request omits its Host header #3077)
• 4b4de3a89 — fix(agent-gaia): stop the npm sidecar attaching to a foreign server (fix(agent-gaia): stop the npm sidecar attaching to a foreign server #3078)
• 6692361d6 — fix(agent-gaia): end a turn parked on a confirmation when stdin closes (fix(agent-gaia): end a turn parked on a confirmation when stdin closes #3080)
• 3ec80f141 — fix(tui): run an installed hub agent over the daemon, not stdio (fix(tui): run an installed hub agent over the daemon, not stdio #3062)
• c7c4cff17 — fix(agent): bill tool-approval time to the human, not the tool (fix(agent): bill tool-approval time to the human, not the tool #3007)
• b388562cc — ci(installers): add job timeouts (ci(installers): add job timeouts #3020)
• c4a4fef42 — fix(ci): the README CI badge is permanently red and reports (fix(ci): the README CI badge is permanently red and reports #3050)
• e2ec50523 — docs(cli): remove duplicate terminal hub section (docs(cli): remove duplicate terminal hub section #3045)
• 215477ab6 — ci: fix stale triggers and timeout left by the agent collapse (ci: fix stale triggers and timeout left by the agent collapse #3067)
• 90c3132db — docs: remove the Blender workshop left dead by the agent collapse (docs: remove the Blender workshop left dead by the agent collapse #3068)
• a4b84fae7 — perf(claude): cache the fixed prefill, and report the cache counters (perf(claude): cache the fixed prefill, and report the cache counters #3026)
• fdd9f18c3 — refactor(agents): collapse per-task agents into flagship skills (refactor(agents): collapse per-task agents into flagship skills #2995)
• 5b4f1381d — fix(ci): scheduled Claude audits were green but never ran (fix(ci): scheduled Claude audits were green but never ran #3053)
• 78bbc3076 — feat(website): TUI-first redesign — download the terminal hub, lead with the flagship agent (feat(website): TUI-first redesign — download the terminal hub, lead with the flagship agent #3030)
• 6faec485e — ci(release): make the TUI binaries reproducible, drop the R2 probe (ci(release): make the TUI binaries reproducible, drop the R2 probe #3040)
• 02691245e — fix(hub): strip R2 credentials, and diagnose why R2 refuses them (fix(hub): strip R2 credentials, and diagnose why R2 refuses them #3037)
• aae4a50ce — ci(hub): require component manifests to match the release tag again (ci(hub): require component manifests to match the release tag again #3036)
• 8ca0f7d98 — fix(hub): give the website redeploy a repo to dispatch against (fix(hub): give the website redeploy a repo to dispatch against #3038)
• 20e081909 — fix(hub): unblock large-artifact publishing, and version components independently (fix(hub): unblock large-artifact publishing, and version components independently #3035)
• 9ad1cee3e — ci(cloud): provision the dev install in Claude Code cloud sessions (ci(cloud): provision the dev install in Claude Code cloud sessions #3033)
• 0befe27ce — ci(linux): install Lemonade from the PPA, not deleted PyPI package (ci(linux): install Lemonade from the PPA, not deleted PyPI package #3034)
• db170b8a2 — fix(agent-gaia): make the flagship agent releasable — RAG in the binary, 0.1.1 on npm (fix(agent-gaia): make the flagship agent releasable — RAG in the binary, 0.1.1 on npm #3031)
• 85ba09230 — fix(hub): stop boto3's default checksums breaking every R2 upload (fix(hub): stop boto3's default checksums breaking every R2 upload #3029)
• 5f3e7bbf1 — feat(hub): publish oversized artifacts direct to R2, bypassing the Worker body cap (feat(hub): publish oversized artifacts direct to R2, bypassing the Worker body cap #3025)
• b254708f1 — ci(hub): one approval per component release, not two (ci(hub): one approval per component release, not two #3018)
• a0ecb0f68 — ci(eval): pin the agent eval to Strix Halo, and record which box measured (ci(eval): pin the agent eval to Strix Halo, and record which box measured #3017)
• 672cd1c5f — fix(init): fail loudly when the Agent UI build can't produce a usable frontend (fix(init): fail loudly when the Agent UI build can't produce a usable frontend #2887)
• 29a87fd87 — ci(eval): capture llama-server's error so the embedder failure is diagnosable (ci(eval): capture llama-server's error so the embedder failure is diagnosable #3015)
• c8987ca67 — fix(email): declare the sidecar bearer gate in the OpenAPI contract (fix(email): declare the sidecar bearer gate in the OpenAPI contract #3002)
• 62e5611d9 — fix(email): "0 messages" for mail that exists — unsupported newer_than: duration unit (fix(email): "0 messages" for mail that exists — unsupported newer_than: duration unit #2981)
• 746468274 — fix(cli): exit non-zero from gaia agent status <id>, add --json (fix(cli): exit non-zero from gaia agent status <id>, add --json #3001)
• 9c89cb472 — ci(hub): deploy the Agent Hub Worker before publishing to it (ci(hub): deploy the Agent Hub Worker before publishing to it #2991)
• 47e6bc908 — fix(gaia): keep a loaded skill alive across turns; stop a control resize from wrecking the screen (fix(gaia): keep a loaded skill alive across turns; stop a control resize from wrecking the screen #2932)
• 9ed3c3523 — test(telegram): isolate HOME so the pid test can't clobber a live adapter (test(telegram): isolate HOME so the pid test can't clobber a live adapter #2895)
• 225c69377 — test(init): stop init tests overwriting the real ~/.gaia/config.json (test(init): stop init tests overwriting the real ~/.gaia/config.json #2891)
• 8ed5b783d — fix(release): emit release notes as short bullets, not prose (fix(release): emit release notes as short bullets, not prose #2967)
• 1beb714f3 — release(agent-email): 0.6.0 (release(agent-email): 0.6.0 #2933)

Full Changelog: v0.23.0...v0.24.0

Release checklist

• util/validate_release_notes.py passes
• docs && npx mintlify validate — could not run mintlify validate locally (npm TLS cert error in this environment), but util/check_doc_links.py passes internal + external link checks now that docs(reference): add a user guide for Claude Code session analytics #3782 is merged
• src/gaia/version.py → 0.24.0; LEMONADE_VERSION unchanged at 11.8.1
• webui package.json / package-lock.json → 0.24.0
• Navbar label → v0.24.0 · Lemonade 11.8.1
• Hub component manifests (terminal-hub, agent-ui) → 0.24.0; check_component_core_api.py --release-version 0.24.0 passes
• hub/agents/gaia/python/{gaia-agent.yaml,pyproject.toml} min-version floor bumped 0.23.1 → 0.24.0 (0.23.1 was never tagged)
• docs/releases/index.mdx corrected: 3 files → 5 (was missing the hub-manifest and webui-package files)
• Rebased onto current main; changelog regenerated from the real range — 230 commits, no phantom SHAs, no missing merges
• Every PR citation and contributor handle verified against gh pr view (33 PRs + 8 contributor logins)
• Full commit range swept for security/breaking markers not caught by title alone
• Review from @kovtcharov-amd addressed

[github-actions]

Skill audit

Skill	Verdict	Claimed tier	Cleared tiers	Findings	Rules
hub/agents/email/npm	✅ ALLOW	experimental	experimental, community	2 medium, 1 info	code.suppression, supply.undeclared_dependency
hub/skills/inbox-triage	✅ ALLOW	community	experimental, community	none	—

✅ All audited skills cleared the tier they claim.

Per-finding detail is withheld here on purpose. Read it in the Security > Code scanning tab, or download the skill-audit-reports artifact from this run. Offending source text is withheld from CI everywhere — reproduce it locally with gaia skill audit <dir> --show-snippets.

[github-actions]

Verdict: Request changes

The v0.24.0 bump itself is clean — every version file agrees and the CLI, the Agent UI backend, and the release-notes validators all confirm 0.24.0. Two problems are in the notes themselves, and one of them turns a CI gate red.

Blocking:

1. The "Session Analytics guide" link in the new notes points at a page that doesn't exist — it returns 404 today, and the guide that would create it is still an open PR. The docs link check runs on every PR touching docs/ and has no tolerance flag, so this fails the gate now and, if tagged as-is, ships a dead link to everyone reading the release. Either land that guide first or drop the link from the bullet.

2. The full changelog is inaccurate in both directions. Four of its entries name commits that exist nowhere in this repository — they're from a branch that was squashed on merge, and two of them are a merge commit and a formatting-only commit that shouldn't appear in a user-facing list at all. Meanwhile two real merged fixes from this window are missing, and the "218 commits" headline doesn't match what actually shipped. Regenerate the list from the real history right before tagging.

Everything else reads well. Deleting the v0.23.1 page is correct — that tag was never cut, so folding its breaking change into these notes is the right call, and the compare link already points at v0.23.0.

Real-world evidence

The automated evidence stage ran on the no-inference lane and exercised every surface this bump touches. It confirms the bump and independently found the broken link above.

$ gaia --version
0.24.0

$ curl -s http://127.0.0.1:4200/api/system/status
HTTP 200
version         = 0.24.0
lemonade_status = None          # expected — no Lemonade on this runner

Version lock (the gate release_components.yml runs on a tag), replicated across all five files: core, both hub/components/*/gaia-agent.yaml, and both webui package files — ALL MATCH at 0.24.0.

$ python util/validate_release_notes.py docs/releases/v0.24.0.mdx --tag v0.24.0
✅ 1 file(s) validated successfully          exit=0

$ python util/check_doc_links.py --internal-only
Total links checked: 987   OK: 987   Broken: 0

$ python util/check_doc_links.py            # includes external URLs
BROKEN LINKS:
  docs/releases/v0.24.0.mdx:28
    URL: https://amd-gaia.ai/docs/reference/session-analytics
    Error: HTTP 404
FAILED: Found broken links                   exit=1

Nav wiring checked out: v0.24.0 present with its file, v0.23.1 removed from both nav and disk together, navbar label v0.24.0 · Lemonade 11.8.1. Spot regression exercised three adjacent operations — /api/health, /api/system/tasks, and a fresh gaia agent init scaffold (stamps 0.24.0 into the generated manifest and pyproject) — all pass. The Agent UI pixels (version string in the app chrome) are marked pending strix-halo lane; per the rubric that deferral is fine for merge-time CI since the underlying route was exercised. The bundle also notes a pre-existing isort failure in the gaia agent init scaffold template, which this PR does not touch.

🔍 Technical details

🔴 Broken external link fails the docs gate and ships a 404 (docs/releases/v0.24.0.mdx:28)

The bullet links https://amd-gaia.ai/docs/reference/session-analytics. That page does not exist locally (docs/reference/session-analytics.mdx is absent, and session-analytics appears nowhere in docs/docs.json), so publishing the site won't create it. It's added by PR #3782, which is still open — so the notes also credit an unmerged PR in a shipped changelog.

.github/workflows/check_doc_links.yml's check-external-links job runs python util/check_doc_links.py --external-only with no continue-on-error, on every PR touching docs/**. Confirmed red by the evidence bundle.

Two options — merge #3782 first (preferred, the guide is real content), or drop the link:

- **Mine your own Claude Code sessions for usage stats** — a deterministic, LLM-free pipeline turns your `~/.claude/projects/` transcripts into tool-usage, failure, and cost reports. (PR [#3111](https://github.com/amd/gaia/pull/3111))

🟡 Four changelog entries reference SHAs that don't exist; two merged PRs are missing (docs/releases/v0.24.0.mdx:171-394)

git cat-file -t reports not found for all four of these, and they're not reachable from any branch — they're pre-squash commits from a PR branch (pr3670):

• 192c39d03 — style: black
• 44a9a6ede — fix(agent): drop whole records from a JSON string, don't slice one
• bc2871849 — Merge remote-tracking branch 'origin/main' into pr3670
• b62eada91 — fix(agent): bound plain-string tool results with head and tail

A merge commit and a style: black commit don't belong in a user-facing changelog regardless of SHA validity.

Missing from the list but present in git log v0.23.0..HEAD:

• ff62b4d5 — fix(agent): strip whitespace from tool names before dispatch (fix(agent): strip whitespace from tool names before dispatch #3744)
• f39ead7a — fix(agent_loop): stop spending hourly budget on idle no-op ticks (fix(agent_loop): stop spending hourly budget on idle no-op ticks #3745)

Count check: 222 commits in v0.23.0..HEAD, minus the 6 release commits in this PR = 216 substantive, against the 218 the notes claim (214 real entries + 4 phantom). Regenerating with git log v0.23.0..<tag-sha> --no-merges immediately before tagging fixes the SHAs, the count, and the omissions in one pass.

🟢 Nits (2)

1. Stale floor comment in the flagship agent package (hub/agents/gaia/python/gaia-agent.yaml:30-31, pyproject.toml:17-31). It pins min_gaia_version: "0.23.1" / amd-gaia>=0.23.1 with a comment saying the floor "only resolves once 0.23.1 is cut" — 0.23.1 was never tagged and never will be. Resolution still works (0.24.0 > 0.23.1), so this is cosmetic, but bumping the floor to 0.24.0 makes the comment true again.

2. The release process doc is now out of date (docs/releases/index.mdx). It states a release PR "must update exactly 3 files", while this PR correctly bumps five — the two hub/components/*/gaia-agent.yaml manifests and the webui package.json / package-lock.json are now gated by release_components.yml too. Worth a one-line fix so the next release doesn't miss them.

Strengths

• The bump is genuinely complete across every gated surface — core, both component manifests, and both webui package files — which is exactly the drift that made v0.23.1 untaggable.
• Deleting docs/releases/v0.23.1.mdx is right, not an oversight: v0.23.1 has no tag and no GitHub release (latest is v0.23.0), so folding its "thirteen per-task agents removed" breaking change into these notes and keeping the v0.23.0...v0.24.0 compare link is the correct history.
• The notes follow the house style well — user-observable impact first, mechanism second, security section leading with what an attacker could actually do.