Significant-Gravitas/AutoGPT autogpt-platform-beta-v0.8.0 on GitHub

⚠️ Upgrading a self-hosted install — breaking change

ENCRYPTION_KEY, UNSUBSCRIBE_SECRET_KEY and BETTER_AUTH_SECRET no longer ship with values in .env.default; they are generated per install, and the backend does not start on the value .env.default used to contain. A fresh install needs nothing beyond make init-env (or the installer). An install that set its own values needs nothing. If your install ran on the defaults (no ENCRYPTION_KEY line in backend/.env, or startup stops with ENCRYPTION_KEY is set to a value that was published…), do this from autogpt_platform/ — it moves your connected integrations to the new key instead of losing them:

docker compose down
Keep the key your data is encrypted with:
export OLD_ENCRYPTION_KEY=$(git show autogpt-platform-beta-v0.7.4:autogpt_platform/backend/.env.default | grep '^ENCRYPTION_KEY=' | cut -d= -f2-)
(if you had set your own key and are replacing it, export that value instead and keep a copy of backend/.env outside the checkout).
In backend/.env make sure ENCRYPTION_KEY= and UNSUBSCRIBE_SECRET_KEY= exist with nothing after the =, and BETTER_AUTH_SECRET= in frontend/.env; then make init-env (fills every empty secret, never overwrites a value).
Re-encrypt what is stored — dry run first, then apply:
docker compose build migrate rest_server && docker compose run --rm migrate
docker compose run --rm --no-deps -e OLD_ENCRYPTION_KEY rest_server cli rotate-encryption-key
docker compose run --rm --no-deps -e OLD_ENCRYPTION_KEY rest_server cli rotate-encryption-key --apply
Clear the token signing key stored under the old BETTER_AUTH_SECRET, then start:
docker compose up -d --wait db
docker compose exec db psql -U postgres -c 'DELETE FROM platform."UserAuthJwks";'
docker compose up -d --build

Full instructions: Upgrading: secrets are generated per install.

Also in this release: the legacy HS256 JWT verification path is removed, so a JWT_VERIFY_KEY in backend/.env is no longer read (#14519); the unused TrialNotificationDelivery table is dropped by migration.

🔥 What's New?

New Features

#14688 - Standing work: routines for experts and Otto, repeating or once (by @ntindle)
#14331 - Let the Discord bot mention server members and roles, never everyone (by @Bentlybro)
#14313 - Dedicated, shareable expert pages with a Coming soon state (by @0ubbe)
#14273 - Expert proposal cards outside the tool chain with approve flow and taglines (by @Abhi1992002)
#14187 - Tag the settings-billing paying surfaces for DataFast (by @ntindle)
#14186 - Report each onboarding step to DataFast (by @ntindle)
#13299 - Alert + gate admin impersonation start via Discord (by @ntindle)
#11791 - Add button to generate data to admin dashboard on local dev (by @ntindle)
#14332 - Connect every provider a tool needs, then proceed once (by @Pwuts)
#14342 - Group Home's Recent work by expert, workflow, or Autopilot (by @0ubbe)
#13261 - Add Linux and macOS appliance installer (by @ntindle)
#14358 - Add GPT-6 Astra support (by @ntindle)
#14385 - Add Opus 5 and upgrade Advanced Anthropic (by @Torantulino)
#14075 - Support Basic authentication for MCP servers (by @Torantulino)
#14403 - Experts first, one-line runs, and run triggers on Home's Recent work (by @0ubbe)
#14361 - Add actionable automation credit history (by @Torantulino)
#14399 - Add expert budget endpoint and run source (by @Abhi1992002)
#14402 - Add Claude Fable 5.1 support (by @ntindle)
#14425 - Expert new-chat button in the sidebar and expert-aware empty session (by @0ubbe)
#14322 - Look Mum No Hands - AutoPilot Voice Mode (by @Pwuts)
#14411 - Experts do the public-data part of the job when a connection is skipped (by @Pwuts)
#14428 - Canonical categories and a category filter (by @Pwuts)
#14353 - Add configurable card-required trials (by @ntindle)
#14422 - Add Google Gemini 3.8 Flash support (by @ntindle)
#14446 - Give marketplace pages real Open Graph link previews (by @Pwuts)
#14431 - List knowledge skills in the marketplace and install them to AutoPilot (by @Pwuts)
#14424 - Let AutoPilot read the user's expert chats (by @Pwuts)
#14244 - Per-tree envelope and ledger for spawned turns (by @Pwuts)
#14420 - Pause expert work for approval once her spend reaches the threshold (by @Pwuts)
#14423 - AutoPilot authors the morning brief (by @Pwuts)
#14432 - Publish knowledge skills to the marketplace and review them (by @Pwuts)
#14395 - Make silent credential-connection failures visible (by @Pwuts)
#14421 - Score each expert's writing style against a stored baseline (by @Pwuts)
#14452 - Update Slant3D integration for API v2 (by @ntindle)
#14460 - Add generated expert avatars across team and chat surfaces (by @Abhi1992002)
#14461 - Choose generated avatars when raising experts (by @Abhi1992002)
#14462 - Recommend experts from onboarding brain dumps (by @Abhi1992002)
#14464 - Introduce an expert-first onboarding wizard (by @Abhi1992002)
#14465 - Introduce the expert team model in copilot onboarding (by @Abhi1992002)
#14466 - Teach AutoPilot to propose the first expert hire (by @Abhi1992002)
#14490 - Show an expert's integrations as logos on the team card (by @0ubbe)
#14037 - Expert profiles show required services, cost, and AI disclosure (by @Abhi1992002)
#14379 - Run copilot sandboxes on our own 1 vCPU / 2 GiB desktop image (by @ntindle)
#14505 - Show what an expert has learned in its Soul drawer (by @Pwuts)
#14512 - Expert profiles show what the expert sets up on day one (by @Pwuts)
#14389 - Add RMFG manufacturing blocks (by @ntindle)
#14506 - Show a bundled workflow's cadence before hire, and skip install-time schedules a hire cannot run (by @Pwuts)
#14504 - Return experts from search and the category chips (by @Pwuts)
#14458 - Add Alibaba Qwen 3.8 Max (0902) support (by @ntindle)
#14414 - Skills owned per expert or personal AutoPilot (by @Abhi1992002)
#14511 - Expert profiles show only the real Skills a hire comes with (by @Pwuts)
#14202 - Add Microsoft 365 Copilot transport (by @ntindle)
#14286 - Record how each agent run started and emit activation events (by @ntindle)
#14287 - Record A/B experiment arms per user (by @ntindle)
#14459 - Add Meta Muse Spark 1.3 support (by @ntindle)
#14530 - Add DeepSeek V4.1 Flash (by @ntindle)
#14393 - AutoPilot says what its task has left and wraps up before it runs out (by @Pwuts)
#14559 - Install a marketplace skill straight onto one of your experts (by @Pwuts)
#14544 - Make a skill package's files reachable and runnable in a turn (by @Pwuts)
#14549 - Store, copy and delete a skill as its whole package (by @Pwuts)
#14555 - Upload and download a skill as a zip package (by @Pwuts)
#14538 - Add Sakana Fugu Ultra v2 (by @ntindle)
#12945 - Add moderation blocks (by @ntindle)
#14578 - Let the chat-platform bot stay silent with NO_REPLY (by @Bentlybro)
#14550 - Raise experts with real marketplace skills (by @0ubbe)
#14557 - Publish and install a skill as its whole package (by @Pwuts)
#14561 - Show a skill's package on the Hub and in admin review (by @Pwuts)
#14543 - Split marketing into four scoped experts (by @0ubbe)
#14415 - Experts run only installed workflows with granted credentials (by @Abhi1992002)
#14416 - AutoPilot manages all schedules, experts see only their runs (by @Abhi1992002)
#14417 - Grant access from the connect card in expert chats (by @Abhi1992002)
#14365 - Interactive E2B desktop sandboxes with live embedded streams (by @ntindle)
#14382 - One box per owner — the desktop is the shell's screen (by @ntindle)
#14288 - One anonymous id across PostHog and LaunchDarkly, stored at signup (by @ntindle)
#14436 - Let AutoPilot edit its own chat-platform messages (by @Bentlybro)
#14514 - Add supported official MCP integrations (by @ntindle)
#14568 - Capability registry core and index (by @ntindle)
#14038 - Experts funnel + briefing instrumentation (by @Abhi1992002)
#14207 - Live sessions message each other, and one cross-expert check (by @Pwuts)
#14639 - Add TypeSafe Jev judgment blocks (by @Torantulino)
#14569 - Swap copilot discovery to find/describe/run_capability (by @ntindle)
#14635 - Add Inception Mercury 2.5 (by @ntindle)
#14675 - Add nine business core experts (by @0ubbe)
#14707 - Show expert job titles on the pill and ask for one when raising (by @0ubbe)
#14677 - Use the credential the user picked, and ask instead of guessing (by @Bentlybro)

Enhancements

#14310 - Count graph runs by terminal status (by @Bentlybro)
#14319 - Wire the application metrics that were defined but never recorded (by @Bentlybro)
#14208 - Upgrade claude-agent-sdk to 0.2.152 and pin the context window explicitly (by @Pwuts)
#14360 - Clarify Max model access in picker (by @Torantulino)
#14398 - Add compact variants to core atoms and restyle Home (by @Abhi1992002)
#14400 - Inline expert chat and team page polish (by @Abhi1992002)
#14404 - Rebuild Files page as a list with hover previews (by @0ubbe)
#14302 - Speak MCP 2026-07-28 alongside legacy servers (by @kcze)
#14408 - Colour Home's work chips and show workflow pictures (by @0ubbe)
#14442 - Recalibrate AutoPilot's per-plan daily and weekly spend limits (by @Pwuts)
#14336 - Shrink the schemas and tool results AutoPilot reads (by @Pwuts)
#14469 - Files multi-select, Team setup card, and Team/Home design polish (by @0ubbe)
#14463 - Refresh brain dump recording with an animated AutoPilot (by @Abhi1992002)
#14497 - Replace the expert kickoff auto-run with an onboarding card (by @0ubbe)
#14571 - Promote GPT-6 Astra to Codex advanced-tier default (by @Otto-AGPT)
#14548 - Notion-style expert avatars with a face picker in the raise flow (by @0ubbe)
#14457 - Build the backend image once instead of eight times (by @Bentlybro)
#14603 - Notify sessions when the skill registry drifts since session start (by @ntindle)

UI/UX Improvements

#14266 - Show error toast when a requested graph fails to load (by @Pwuts)
#14368 - Calm the Recent work card on Home (by @0ubbe)
#14407 - Use solid expert colors for covers (by @Abhi1992002)
#14300 - Ignore Enter and other keys while an IME is composing (by @kcze)
#14472 - Hide unfinished expert pods UI (by @Abhi1992002)
#14437 - Format sidebar date group labels with the user's locale (by @gaoanze888)
#14673 - Move billing toggle closer to plan cards in onboarding (by @0ubbe)
#14661 - Unify expert display and sidebar chats (by @0ubbe)
#14696 - Put the desktop back in reach of the chat (by @kcze)
#14700 - Restore result cards for tools run through run_capability (by @kcze)
#14682 - Stopping a response shows the stopped card, not an error (by @Pwuts)

Bug Fixes

#14721 - Apply the building-mode guide on restart instead of re-deriving it from history (by @Pwuts)
#14714 - Run a tool reached by capability id through the normal tool path — AutoPilot can create and edit agents again (by @Pwuts)
#14702 - Name deferred tools by capability id in model-facing text (by @kcze)
#14330 - Read the post a Discord thread was created from (by @Bentlybro)
#14325 - Skip scheduled runs for paywalled users quietly (by @Bentlybro)
#14337 - Show expert cards to signed-out visitors (by @0ubbe)
#14324 - Route every Stripe SDK call through one async helper (by @Bentlybro)
#14334 - Upgrade redis-py to 8.1 so a dropped cluster connection no longer poisons the client (by @Bentlybro)
#14057 - Render array items with a known object shape as a real sub-form (by @ntindle)
#14206 - Raise the LLM request timeout to 600s and make it configurable (by @Pwuts)
#14327 - Stop managed, stale and trigger credentials breaking the Connect card (by @Pwuts)
#14212 - Show what the approval card will execute, and approve only that (by @Pwuts)
#14320 - Invalidate the copilot integration-token cache across processes (by @Pwuts)
#14309 - A sub-session hands back a bounded result, not its whole tool log (by @Pwuts)
#14205 - Require library membership AND marketplace submission for graph access (by @Pwuts)
#14373 - Render canonical agent and block names in AutoPilot (by @Torantulino)
#13524 - Authorize team management by target team id, drop active-team requirement (by @ntindle)
#14394 - Match sensitive header names case-insensitively when following redirects (by @Pwuts)
#14441 - Drop first-person voice from run activity summaries (by @0ubbe)
#14303 - Stop deleting valid MCP credentials on a 403, and verify tokens server-side (by @kcze)
#14445 - Gate the expert page's Coming soon on the hire-experts flag (by @0ubbe)
#14419 - Stop replaying Authorization across cross-origin redirects (by @kcze)
#14426 - Keep each expert's workspace files separate from other experts (by @0ubbe)
#14451 - Bind the scheduler's RPC port before the embedding backfill (by @Pwuts)
#14412 - Facts an expert stores now persist as recallable memories (by @Pwuts)
#14448 - Keep block and agent inputs off the copilot setup card (by @0ubbe)
#14434 - Use the timezone AutoPilot was given when scheduling an agent (by @Pwuts)
#14390 - Tell AutoPilot and the user when a stored credential is rejected, not "not connected" (by @Pwuts)
#14454 - Seed E2E embeddings before CI dumps the database for its cache (by @Pwuts)
#14392 - Tag Vercel previews as preview in Sentry, and stop pytest runs reporting (by @Pwuts)
#14391 - Store the scopes the provider granted, not the ones we asked for (by @Pwuts)
#14375 - Never drop a user's tier on an unauthoritative payment-flag read (by @Pwuts)
#14374 - Cap the compaction window pin at the routed model's real window (by @Pwuts)
#14354 - Read aloud falls back to our own TTS when the browser has no voices (by @Pwuts)
#13420 - Use custom rjsf validator for anyOf fields (by @seer-by-sentry[bot])
#13576 - Base64-encode the ExecuteCodeBlock variables payload (by @albatrossflyon-coder)
#14372 - Isolate copilot dispatch sleep mocks (by @Torantulino)
#14388 - Silence AbortError in onboarding provider (by @seer-by-sentry[bot])
#14323 - Guard Tally popup script removal to prevent NotFoundError (by @seer-by-sentry[bot])
#14453 - Prevent agent page crash on 404 (by @seer-by-sentry[bot])
#14260 - Guard AppSidebar keydown listener against undefined event.key (by @seer-by-sentry[bot])
#14179 - Prevent Tooltip from switching between controlled/uncontrolled (by @seer-by-sentry[bot])
#13498 - Prevent ApiError [object Object] from FastAPI 422 detail (by @seer-by-sentry[bot])
#13581 - Guard against undefined platform in PlatformBadge (by @seer-by-sentry[bot])
#13166 - Remove benign console.warn from credential input (by @seer-by-sentry[bot])
#14491 - Stop the hire kickoff from telling the expert to run a workflow (by @Pwuts)
#14492 - Stop logging Ayrshare requests (by @Pwuts)
#14500 - Refresh every expert list after connecting a service (by @0ubbe)
#13356 - Trim Ayrshare profile keys in request headers (by @pony-maggie)
#14515 - Send question answers right away and advance on option pick (by @Abhi1992002)
#14513 - Stop the roster seed from overwriting hired experts' skills (by @Pwuts)
#14178 - Stop a tool result rendering its own discriminator (by @ntindle)
#14177 - Stop device auth reporting a new credential as removed (by @ntindle)
#14326 - Avoid retrying Redis during test collection (by @Ftgn-dpA)
#14095 - Let device-code providers connect from copilot chat (by @ntindle)
#14517 - Disable CoPilot spend caps for self-hosted installs (by @ntindle)
#14524 - Store marketplace media locally when GCS is unset (by @ntindle)
#14526 - Distinguish experts from Otto in UI and tools (by @Torantulino)
#14529 - Await the onImagesChange assertion in ThumbnailImages local-media test (by @Pwuts)
#14486 - Strip whitespace in bool string conversion (by @soroush5)
#14488 - Wrap strings as single element in set/tuple conversion (by @soroush5)
#14547 - Keep a hire's onboarding card on the expert's own role (by @0ubbe)
#13750 - Stabilize ReactFlow prop references on the builder canvas (by @chelsealong)
#14439 - Stop unpickling paused/fired schedules on every get_execution_schedules read (by @Bentlybro)
#14542 - Reap the browser process tree when a turn ends (by @Bentlybro)
#14556 - Test fixtures delete the templates and hires they create (by @Pwuts)
#14435 - Deliver AutoPilot clarifying questions to bot platforms (by @Bentlybro)
#14501 - Make the bot's mentions actually ping, by name or id (by @Bentlybro)
#14468 - Block the CLI cron tools that silently schedule nothing (by @kcze)
#14363 - Enforce string truncation size limits (by @QiuLsG)
#13527 - Gate org credits routes behind MANAGE_BILLING (by @ntindle)
#14618 - Give a second card the credential the expert was granted (by @Pwuts)
#14621 - Make Remy a skills-only expert (by @0ubbe)
#14519 - Remove legacy HS256 verification path (by @kcze)
#14614 - Align autogpt_libs PyJWT constraint with the backend (by @kcze)
#14627 - Await the Add buttons in the MarketplaceStep test (by @Pwuts)
#14628 - Keep the copilot system prompt fresh after a Langfuse publish (by @Pwuts)
#14631 - Keep acronym casing in marketplace skill titles (by @Pwuts)
#14634 - Hide the pooled org balance on /home from members without MANAGE_BILLING (by @Pwuts)
#14633 - Make the second-opener-waits desktop test deterministic (by @Pwuts)
#14641 - Stop the expert-kickoff tests inheriting a stale kickoff latch (by @Pwuts)
#14218 - Park unrestorable scheduler jobs instead of deleting them (by @Pwuts)
#14643 - Prevent localStorage errors on restricted WebViews (by @sentry[bot])
#14518 - Generate local secrets at bootstrap instead of shipping them (by @kcze)
#14670 - Desktop stream fails to start since the VNC stack runs as root (by @kcze)
#14667 - Offer the reconnect card when a credential cannot be refreshed (by @Bentlybro)
#14662 - Give the sandbox the credential the connect card shows as connected (by @Bentlybro)
#14686 - Queue a message on a plain-HTTP origin without crashing the chat (by @Bentlybro)
#14687 - Read the host from a URL whose scheme is uppercase (by @Pwuts)
#14689 - Stop reporting a user's cancel as an assistant error (by @Pwuts)
#14679 - Let users enter host-scoped and username/password credentials (by @Pwuts)
#14709 - Allow skipping expert job titles (by @0ubbe)
#14698 - Restart the desktop stream when x11vnc died behind a live noVNC (by @kcze)
#14699 - Stop the desktop stream when the box pauses (by @kcze)
#14684 - Deliver a Stop to every executor pod, not one of them (by @Pwuts)
#14685 - Deliver a graph-run cancel to every pod, not one of them (by @Pwuts)
#14710 - Keep a non-default port on a host-scoped credential's host (by @Pwuts)
#14712 - Generate secrets in the Compose installer and add an upgrade path (by @kcze)

Documentation

#14335 - Publish new SECURITY.md (by @Pwuts)
#14305 - Post pr-test screenshots as comment attachments, not repo branches (by @Pwuts)
#14387 - Point the Forge quickstart link at a file that exists (by @firekou)
#14470 - Regenerate block docs for Gemini 3.8 Flash (#14422 skipped it) (by @Pwuts)
#14483 - Regenerate block docs for the AutoPilot tool names #14424 added (by @Pwuts)
#13998 - Fix dead feature-requests link in README (by @putramkti)
#14236 - Refresh agent guidance paths (by @Ricky-7-Yan)
#14447 - Refresh manual guides and links (by @ntindle)
#13048 - Document integration icon, auth patterns, auto-discovery, docs sync, and test patterns (by @ntindle)
#14539 - Align AGENTS.md matcher with suite (no jest-dom) (by @ntindle)

Dependencies & Maintenance

#14720 - Bump anyio to 4.15.1 for CVE-2026-63374 (by @Pwuts)
#14348 - Drop the trailing blank line black rejects in org_credit_test (by @Bentlybro)
#14366 - Record agent and model attribution (by @Torantulino)
#14396 - Remove the four dormant claude-code-action workflows (by @Pwuts)
#14450 - Name the icon library the frontend actually uses (by @Pwuts)
#14455 - Widen the AutoPilot tool-schema budget to 65,000 chars (by @Pwuts)
#14484 - Raise the AutoPilot tool-schema budget to 67,651 chars for the in-flight expert PRs (by @Pwuts)
#14495 - Refresh the expert writing-style baseline (by @Pwuts)
#14278 - Make Classic Python checks dispatch safe (by @ntindle)
#14067 - Document the GitHub review bot as an option in PR skills (by @ntindle)
#14277 - Register scored workflow dispatches (by @ntindle)
#14279 - Shard tests with fail-closed reports (by @ntindle)
#14536 - Reject stale skip allowances across shard reports (by @ntindle)
#14280 - Shard integration tests with verified coverage (by @ntindle)
#14502 - Read and write block docs as UTF-8 regardless of platform (by @Bentlybro)
#14606 - Let secret-gated skips pass on fork PRs (by @Pwuts)
#14630 - Seed a fixture roster in tests, never the real one (by @Pwuts)
#14632 - Credits authz invariants see the whole app and assert instead of exempt (by @Pwuts)
#14644 - The payment-method route opens the caller's own billing portal (by @Pwuts)
#14647 - Fix what stops Dependabot PRs from opening or passing CI (by @kcze)
#14651 - Bump @chromatic-com/storybook from 4.1.2 to 4.1.3 in /autogpt_platform/frontend (by @dependabot[bot])
#13274 - Bump the development-dependencies group across 1 directory with 3 updates (by @dependabot[bot])
#14562 - Bump the production-dependencies group across 1 directory with 43 updates (by @dependabot[bot])
#13277 - Bump cryptography from 46.0.5 to 50.0.1 in /autogpt_platform/autogpt_libs (by @dependabot[bot])
#13279 - Bump faker from 38.3.0 to 40.39.0 in /autogpt_platform/backend (by @dependabot[bot])
#12038 - Bump groq from 0.30.0 to 1.7.0 in /autogpt_platform/backend (by @dependabot[bot])
#12045 - Bump aiodns from 3.6.1 to 4.0.4 in /autogpt_platform/backend (by @dependabot[bot])

Refactoring

#14343 - Refine team and expert pages (by @Abhi1992002)
#14210 - Narrow the external-action soul rule to what is enforced (by @Pwuts)
#14456 - Move the API-key routes into their own feature module (by @Pwuts)
#14471 - Move the copilot-skills routes into their own feature module (by @Pwuts)
#14473 - Move the schedule routes into their own feature module (by @Pwuts)
#14474 - Move the execution routes into features/graph_executions (by @Pwuts)
#14475 - Move the graph routes into their own feature module (by @Pwuts)
#14476 - Move the credit ledger routes into features/billing/credits (by @Pwuts)
#14477 - Move the subscription and Stripe routes into features/billing/subscriptions (by @Pwuts)
#14478 - Move the block and file-upload routes into features/blocks (by @Pwuts)
#14629 - Stop declaring tools an automation session cannot call (by @Pwuts)
#14479 - Move the onboarding routes into features/onboarding (by @Pwuts)
#14480 - Move the user-settings routes into features/user and delete v1.py (by @Pwuts)
#14481 - Give the four colliding response models unambiguous names (by @Pwuts)
#14637 - Match the trial card to the plan cards and billing cards (by @0ubbe)
#14648 - Delete find_block, the tool nothing calls (by @ntindle)

🎉 Thanks to Our Contributors!

A huge shoutout to everyone who contributed to this release. Special thanks to our new contributors:

As well as our returning contributors:

📥 How to Get This Update

To update to this version, run:

git pull origin autogpt-platform-beta-v0.8.0

Or download it directly from the
Releases page.

For a complete list of changes, see the Full Changelog.

📝 Feedback and Issues

If you encounter any issues or have suggestions, please join our Discord and let us know!

Significant-Gravitas/AutoGPT autogpt-platform-beta-v0.8.0 AutoGPT Platform v0.8.0 on GitHub