feat(platform): add single-container distribution - #13758
Conversation
github-project-automation
Bot
moved this to 🆕 Needs initial review
in AutoGPT development kanban
|
Note Reviews pausedIt looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the Use the following commands to manage reviews:
Use the checkboxes below for quick actions:
WalkthroughAdds a complete single-container appliance with bundled services, persistent configuration, supervised startup, local Nginx routing, health probes, recovery handling, administrative tooling, and multi-architecture CI smoke tests. ChangesSingle-container appliance
Estimated code review effort: 5 (Critical) | ~120 minutes Possibly related PRs
Suggested labels: Suggested reviewers: Poem
🚥 Pre-merge checks | ✅ 4 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (4 passed)
✨ Finishing Touches 💡 1📝 Generate docstrings 💡
🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## dev #13758 +/- ##
==========================================
- Coverage 77.95% 77.93% -0.02%
==========================================
Files 2921 2921
Lines 219700 219693 -7
Branches 20836 20835 -1
==========================================
- Hits 171258 171216 -42
- Misses 43840 43871 +31
- Partials 4602 4606 +4
Flags with carried forward coverage won't be shown. Click here to find out more.
🚀 New features to boost your workflow:
|
PR #13758 single-container test reportVerdict: BLOCKED — runtime acceptance could not begin because the image does not build on the available Docker storage stack. Results
Build blockerBoth independent VMs use Docker with A three-line Dockerfile containing only the same pinned remote The primary VM additionally reproduced unsupported hashing on cross-stage RecommendationDo not treat PR #13758 as runtime-verified from this run. Run the architecture smoke workflow from stacked PR #13759 on standard GitHub-hosted Docker workers, or make the Dockerfile inputs compatible with |
…-Gravitas#13755) ### Why / What / How **Why** The single-container appliance supervises backend roles in one network namespace. Internal metrics listeners should not be exposed beyond container loopback, and notification workers must stop cleanly when Supervisor terminates the process. **What** - Binds executor and Copilot executor metrics endpoints to loopback. - Gives notification queue polling an explicit task lifecycle. - Adds regression coverage for clean cancellation and shutdown. **How** The metrics servers use an explicit loopback host. The notification runner tracks its polling task, cancels and awaits it during shutdown, and avoids leaving an event-loop-bound task behind. ### Stack Part **1 of 6** in the single-container stack. This is the bottom PR and targets `dev`. Review and merge the stack bottom-up. It supersedes the corresponding backend portion of draft Significant-Gravitas#13754. ### Changes 🏗️ - 4 focused backend source/test files. - No Docker image or configuration changes in this layer. ### Checklist 📋 #### For code changes: - [x] I have clearly listed my changes in the PR description - [x] I have made a test plan - [x] Signed commit passed the repository pre-commit suite under Node 24, including backend lint, formatting, type checking, API generation, and secret checks - [x] Notification regression coverage is included - [ ] Let PR CI validate this isolated layer #### For configuration changes: - [x] Not applicable; this layer does not change configuration ### Stack navigation Bottom → top: 1. [Significant-Gravitas#13755 — backend supervision](Significant-Gravitas#13755) 2. [Significant-Gravitas#13757 — frontend routing](Significant-Gravitas#13757) 3. [Significant-Gravitas#13756 — runtime options](Significant-Gravitas#13756) 4. [Significant-Gravitas#13758 — single-container image](Significant-Gravitas#13758) 5. [Significant-Gravitas#13759 — validation and publication CI](Significant-Gravitas#13759) 6. [Significant-Gravitas#13760 — operations documentation](Significant-Gravitas#13760) The base refs already form GitHub's required linear chain. The native Stack association is pending GitHub's repository-by-repository public-preview rollout; the Stack API currently returns its documented not-enabled response for this repository. <!-- CURSOR_SUMMARY --> --- > [!NOTE] > **Medium Risk** > Changes shutdown ordering and timeouts for the notification RabbitMQ path and alters metrics bind behavior; mistakes could hang supervisor restarts or affect scrape access, though behavior is heavily tested and bounded. > > **Overview** > Hardens backend processes for **single-container supervision**: internal Prometheus listeners follow the same bind address as other service RPC (`pyro_host`), and the notification service shuts down RabbitMQ consumers with **bounded, non-blocking** teardown. > > **Graph and Copilot executors** now pass `addr=settings.config.pyro_host` into `start_http_server`, so metrics stay on loopback in the appliance runtime while docker-compose can still scrape via `0.0.0.0`. > > **NotificationManager** keeps strong references to the background `_run_service` future/task, routes shutdown through a new `_shutdown_service` barrier (cancel consumers, then disconnect with per-stage timeouts), and replaces the old synchronous disconnect in `cleanup()` with `run_and_wait(..., timeout=CLEANUP_TIMEOUT_SECONDS)` plus handling for already-closed or racing event loops. > > **AppService** gains an optional timeout on `run_and_wait` and skips scheduling `loop.stop()` when the shared loop is already closed. > > Regression tests cover future retention, ordered shutdown, cancellation timeouts, and loop-edge cases. > > <sup>Reviewed by [Cursor Bugbot](https://cursor.com/bugbot) for commit c38bd9e. Bugbot is set up for automated code reviews on this repo. Configure [here](https://www.cursor.com/dashboard/bugbot).</sup> <!-- /CURSOR_SUMMARY --> --------- Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
…into codex/pr13756-merge
…into codex/pr13756-polish
…thub.com/Significant-Gravitas/AutoGPT into codex/pr13758-polish
![autogpt-pr-reviewer[bot]](https://avatars.githubusercontent.com/u/130738209?s=60&v=4){kind=small}
There was a problem hiding this comment.
📋 Automated Review — PR #13758
PR #13758 — feat(platform): add single-container distribution
Author: ntindle | Files: 46
🎯 Verdict: APPROVE
PR Description Quality
✅ Has Why + What + How — the description explains the appliance goal (one docker run to boot the full platform), enumerates the bundled services, and documents first-boot secret generation, /data persistence, and the closed-by-default signup posture.
What This PR Does
Adds a fully self-contained, opt-in single-container distribution of the AutoGPT Platform: one image bundling the frontend, backend services, Postgres, a Valkey cluster, RabbitMQ, FalkorDB, nginx, and Supervisor, runnable with a single docker run. It generates installation secrets on first boot, persists state under /data, exposes only nginx on :3000, and enforces least-privilege isolation (per-service OS users, a scoped autogpt_frontend Postgres role, loopback-pinned internal topology). The change is entirely additive — it lives under autogpt_platform/single-container/ plus one CI workflow and one smoke script, and touches no existing backend/frontend/compose sources.
Specialist Findings
🛡️ Security ✅ — Traced trust boundaries end-to-end (secret generation, per-service users, least-privilege PG role, loopback pinning, privilege-dropping launchers, nginx edge, log redaction); injection surfaces are clean (promote-admin.sh uses psql :'target_email' quoting, frontend role uses passwordless peer auth). No critical/high/medium findings. Two 🔵 defense-in-depth notes only.
🏗️ Architecture ✅ — Cleanly isolated and additive; Bake reuses the backend server target as a named context rather than forking build logic. Main long-term risk is several hand-maintained config mirrors (common.sh:73, entrypoint.sh:155, run-frontend.sh:10, healthcheck.sh:21) that can silently drift. All 🟠/🔵.
⚡ Performance ✅ — Infra-only, no hot request path. Steady-state note: the watchdog (watchdog.sh:25) and Docker HEALTHCHECK (Dockerfile:248) both run the full ~12-HTTP-probe battery every 30s, doubling idle health-probe cost; plus a 3-node Valkey cluster with zero replicas on one host (entrypoint.sh:270).
🟠 Duplicate continuous health probing (watchdog.sh:25)
🧪 Testing ✅ — Exceptionally well-tested: 59 unit tests + an 850-line negative-case smoke script (DB isolation, topology pinning, log-leak prevention, corrupt-config rejection, secret-rotation refusal). Gap: the AUTOGPT_ENABLE_BOT_SERVICES=true path is never exercised.
🟠 Bot-services enabled path untested (.github/scripts/platform-single-container-smoke.sh:795)
📖 Quality ✅ — Readability grade A; descriptive naming, small single-purpose functions, set -Eeuo pipefail throughout, high-signal comments. Nits are DRY loops and the hardcoded healthcheck program roster.
📦 Product ✅ — Requirements fully met, scope disciplined. Two operator-experience concerns: first-admin onboarding requires a multi-step CLI ritual with two container recreations, and AUTOGPT_PUBLIC_URL mismatches fail silently at auth time.
🟠 First-run admin onboarding friction (.env.example:11); 🟠 silent AUTOGPT_PUBLIC_URL footgun (entrypoint.sh:102)
📬 Discussion ✅ — All 266 review threads resolved, GitHub reviewDecision: APPROVED (human approval by Torantulino after the final commit), no merge conflicts (MERGEABLE/CLEAN). Prior sentry/coderabbit/cursor concerns all addressed in landed commits. The stale capy-ai BLOCKED verdict was a harness limitation on an old SHA, superseded.
🔎 QA ✅ — Ran 8 Bash checks: 59 appliance unit tests pass, all shell scripts pass bash -n, bake config resolves, first-boot secret generation (11 keys, mode 0600) and idempotent persistence confirmed, and negative tests passed (4 malformed public-URLs rejected rc=2; env-supplied POSTGRES_PASSWORD rotation refused). Honest limitation: the full appliance image could not be built in-sandbox, so end-to-end docker run behaviors rely on the native amd64/arm64 CI smoke job.
🟠 Should Fix
- Bot-services enabled path untested (
.github/scripts/platform-single-container-smoke.sh:795) — no test setsAUTOGPT_ENABLE_BOT_SERVICES=true, so startup ofcopilot-bot/platform-linking-manager/copilot-chat-bridgeand their health URLs is never exercised; a regression there passes all tests. (Flagged by: testing) - Duplicate continuous health probing (
watchdog.sh:25/Dockerfile:248) — the same full probe battery runs twice per 30s window forever on an appliance meant to sit idle cheaply. Have the watchdog consult Docker health state or a cached marker, or stagger the intervals. (Flagged by: performance) - Healthcheck program roster duplicates supervisord.conf (
healthcheck.sh:21) — hardcoded Supervisor program list silently narrows coverage when a program is renamed/added. Derive fromsupervisorctl statusor add a CI assertion. (Flagged by: quality, architect — 2 specialists) - First-run admin onboarding friction (
.env.example:11) — the advertised singledocker runlands on a login page the operator cannot register on; creating the first admin needs enable-signup → recreate → sign up →docker exec ... promote→ disable → recreate. Surface the bootstrap steps in the disabled-signup UI state or ship a one-time first-run token. (Flagged by: product) - Silent
AUTOGPT_PUBLIC_URLfootgun (entrypoint.sh:102) — defaults tohttp://localhost:3000; validation checks URL validity but not whether it matches the published port, so-p 8080:3000without setting it yields a "healthy" container with broken auth cookies/redirects. Warn loudly at startup when the public URL likely differs from the mapping. (Flagged by: product)
🟡 Nice to Have
- Runtime secret-key allowlist is a third copy (
common.sh:73) — duplicatesruntime_config.py_new_values/_validate_values; add a CI assertion the three sets match. (architect) APP_ENV=devmasquerade (entrypoint.sh:155) — a self-hosted appliance is semantically LOCAL; future dev-gated backend branches could silently activate. PreferLOCAL+ explicit doc/metrics-disable toggle. (architect)- 3-node Valkey cluster with zero replicas on one host (
entrypoint.sh:270) — no HA benefit, 3× process/AOF/gossip cost; use a single instance if the backend allows, else document the cost. (performance) - Static assets proxied through Node (
nginx.conf:117) — serve immutable/_next/static/*directly from nginx viaalias+ long-lived cache headers. (performance) - Belt-and-suspenders gating of openapi/metrics (
nginx/nginx.conf:90) — backend still emits/openapi.jsonand/metrics; a single case-sensitive nginxreturn 404is the sole gate. Also disableopenapi_url/gate/metricsat the app. (security) - CORS origins not pinned to public URL (
entrypoint.sh:155) — exportBACKEND_CORS_ALLOW_ORIGINS="${AUTOGPT_PUBLIC_URL}"; largely inert same-origin but latent if published at a non-localhost origin. (security) - Watchdog fail-stop counter reset untested (
watchdog.sh:31),probe_tcpappears dead / amqp branches untested (probe.py:61),run-service.shunknown-service branch untested (run-service.sh:26). (testing)
🔵 Nits
- DRY loops for Valkey/nginx dirs (
entrypoint.sh:49,:66;bootstrap.sh:27) — collapse repeatedinstall -d/wait_untillines into loops over the port/subdir set. (quality) - Long
|-delimited key allowlist (common.sh:62) — a multi-line array reads and diffs better. (quality) - Add an orientation
README.mdfor the ~15-filesingle-container/directory. (quality)
QA Screenshots
| Screenshot | Description |
|---|---|
 |
Frontend baseline renders (212 KB), confirming the sandbox environment is healthy ✅ |
Human Review Needed
NO — Not required because all changed paths are confined to the isolated AutoGPT single-container distribution; security review completed with no high or critical findings or blockers.
Risk Assessment
Merge risk: LOW | Rollback: EASY (fully additive, isolated directory + one CI workflow; revert removes the feature with no impact on existing deployments)
CI Status
GitHub CI (per discussion review): ✅ ~40/40 checks green on linux/amd64 and linux/arm64 (build+smoke+scan, CodeQL, Codecov gates, lint, types, e2e), reviewDecision: APPROVED, no merge conflicts. Local harness: ✅ all 5 configured checks pass (frontend lint, backend lint, frontend typecheck, frontend unit tests, frontend build).
UI Testing — Variant Results
✅ local: All 59 appliance unit tests, shell syntax/lint, bake validation, first-boot secret generation/persistence, public-URL validation, and secret-rotation negative tests pass; full-image E2E is out of scope for this sandbox and is covered by the native CI smoke job.
✅ hosted: All in-session verifiable checks pass (59/59 unit tests, bash -n on 12 scripts, shellcheck -x rc=0, valid bake plan, all 21 COPY sources present, healthcheck/supervisor names consistent); full image build + smoke test and the additive-only diff claim are CI-gated and could not be reproduced in this sandbox, but no defect was observed.
c749f91
|
🔄 Auto-redeploying: new commits pushed to a PR with an active deployment. Refreshing development environment for PR #13758. |
1 similar comment
|
🔄 Auto-redeploying: new commits pushed to a PR with an active deployment. Refreshing development environment for PR #13758. |
|
/review |
|
🔄 Auto-redeploying: new commits pushed to a PR with an active deployment. Refreshing development environment for PR #13758. |
github-actions
Bot
added
cla: pending
|
/reapprove |
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
There was a problem hiding this comment.
Re-approved at the request of @ntindle (#13758 (comment))
There was a problem hiding this comment.
📋 Automated Review — PR #13758
PR #13758 — feat(platform): add single-container distribution
Author: ntindle | Files: 46
🎯 Verdict: APPROVE
PR Description Quality
✅ Has Why + What + How — the PR explains the goal (a one-docker run local appliance), what it adds (self-contained image bundling backend, frontend, Postgres, Valkey cluster, RabbitMQ, FalkorDB behind nginx), and how (a Bake target layered on the existing backend server image), with a fully-ticked test plan.
What This PR Does
Adds a new, fully self-contained single-container distribution of the AutoGPT platform so it can be run locally as a complete appliance. Everything lives under autogpt_platform/single-container/ plus one CI workflow and one smoke script — it does not touch existing backend/frontend source, Dockerfiles, or Compose, so hosted deployments are unaffected. The image bundles all services with strong per-service OS-user isolation, a least-privilege frontend DB role, generated-once persisted secrets, and an nginx front door that is the only exposed port.
Specialist Findings
🛡️ Security ✅ — Reviewed every runtime script, nginx/supervisor/rabbitmq config, secret generation, and privilege-dropping logic; found no injection, auth-bypass, secret-exposure, or privilege-escalation vulnerability. Verified SQL parameterization in promote-admin.sh:20, 0600 root:root secret files, setpriv cap-dropping (run-frontend.sh:76), host-header/redirect pinning (nginx.conf:120-131), and TOCTOU-safe secret persistence.
🔵 Two informational notes: provider API keys are passed into the frontend env (run-frontend.sh:31, by design for transcription), and the bootstrap-admin window opens unauthenticated signup while enabled (documented, defaults closed).
🏗️ Architecture ✅ — Cleanly isolated and correctly layered on the existing backend image via a named Bake context, with zero coupling introduced into the hosted path. Main risk is unenforced drift between the appliance and backend conventions.
🟠 Supervised-program list is hand-duplicated between healthcheck.sh:21 and supervisord.conf — a missed edit silently drops health coverage or fails every check.
⚡ Performance ✅ — No application hot paths; nginx is well-tuned and probe.py parallelizes probes. Flags steady-state resource tuning for the appliance profile.
🟠 PostgreSQL max_connections left at default 100 (entrypoint.sh:113) against an aggregate Prisma + frontend-role + scheduler pool ceiling that can approach it under concurrent load.
🔵 Redundant dual 30s health suites (Docker HEALTHCHECK + watchdog.sh:43) and AOF on all three Valkey nodes (entrypoint.sh:269) add avoidable steady-state overhead.
🧪 Testing ✅ — Unusually strong: 59 Python unit tests + an 850-line CI smoke test doing real runtime verification (UID isolation, Postgres privilege introspection, restart recovery), with well-represented negative cases.
🟠 Valkey cluster non-fresh recovery / "refusing to erase persistent state" data-safety branch (bootstrap.sh:92) is never exercised.
🔵 Bash config-loader rejection paths (common.sh:77) and the initial-health fail-stop (watchdog.sh:60) untested; some test_entrypoint.py tests grep source substrings rather than behavior.
📖 Quality ✅ — Readability grade A: descriptive naming, why-not-what comments, guard-clause style, atomic writes. Findings are minor DRY/formatting nits (nginx.conf:73 duplicate ws block, infra ports as bare literals in common.sh:12, DSN duplicated in run-frontend.sh:8).
📦 Product docker run" story starts the stack, but signup is disabled by default (AUTH_ALLOW_NEW_ACCOUNTS=false), so first login requires an env toggle, two container recreations, and a CLI promote step documented only in logs/.env.example.
🟠 First-run admin bootstrap is a cliff (.env.example:16) — opt-in/experimental, so not a hard block, but worth smoothing before end-user positioning.
📬 Discussion dev, so reviewDecision is currently REVIEW_REQUIRED; the core Build, smoke, and scan (amd64/arm64) + e2e/python-analyze CI jobs were still pending on head 2d36895.
🔎 QA ✅ — Could not build/boot the multi-GB appliance image in-sandbox (honest limitation), so executed the PR's actual code paths directly: 59 unit tests pass, secret generation is idempotent (259a3375…41c3 stable across re-runs), all 12 shell scripts pass bash -n, and probe.py correctly distinguishes 200 / 404 / connection-refused. All negative tests rejected correctly. Container-integration proof is deferred to CI.
QA Screenshots
| Screenshot | Description |
|---|---|
|
Baseline dev frontend renders (PR does not alter UI) ✅ |
🟠 Should Fix
- Health-check program list duplicated from supervisord (
healthcheck.sh:21) — derive the required-program set fromsupervisorctl statusor add a test asserting the two lists stay in sync. (Flagged by: architect — 1) - PostgreSQL
max_connectionsuntuned vs aggregate pool budget (entrypoint.sh:113) — raisemax_connectionsin the appended config or derive per-serviceconnection_limitfrom a documented total cap, so concurrent load can't surface as connection-refused outages. (Flagged by: performance — 1) - Valkey non-fresh recovery / data-safety branch untested (
bootstrap.sh:92) — add a test that seeds an unrecoverable cluster state and asserts it fails fatally without destroying persisted memory/queue state, plus an already-formed-cluster recovery case. (Flagged by: testing — 1) - First-run admin bootstrap requires two container recreations (
.env.example:16) — provide a one-shotcreate-first-adminpath that doesn't require enabling open signup, and/or surface the bootstrap instructions on the login page instead of only in logs/.env.example. (Flagged by: product — 1)
🟡 Nice to Have
- Schema-drift pointer for frontend DB grants (
bootstrap.sh:170) — comment linking the grant block toschema.prisma/Better Auth so schema changes prompt review here. (architect) - Reduce steady-state overhead (
watchdog.sh:43,entrypoint.sh:269) — let the watchdog read Docker's health status instead of re-running the full suite, and disable/relax AOF on cache Valkey nodes if durability isn't required. (performance) - Cover bash config-loader + initial-health fail-stop branches (
common.sh:77,watchdog.sh:60) and make substring-basedtest_entrypoint.pychecks behavioral where feasible. (testing) - Centralize infrastructure ports (
common.sh:12) — hoist Postgres/Valkey/FalkorDB/frontend ports alongside the app-port constants. (quality)
🔵 Nits
- Duplicate nginx ws location block (
nginx.conf:73) — collapse intolocation ~ ^/_agpt/ws/?$. (quality) - Duplicated frontend DSN literal (
run-frontend.sh:8& Dockerfile) — add a cross-referencing comment or derive from one source. (quality) user=rooton[program:next](supervisord.conf:269) — one-line comment noting root is required for thesetprivcap-drop chain, so a maintainer doesn't "fix" it. (architect, quality)- Three blank lines between supervisor blocks (
supervisord.conf:109) — reduce to one. (quality)
Human Review Needed
NO — Not required because all changed paths are confined to the isolated AutoGPT single-container distribution; security review completed with no high or critical findings or blockers.
Risk Assessment
Merge risk: LOW | Rollback: EASY — the change is fully additive and opt-in; deleting the single-container/ directory and its two CI/smoke files reverts it with no impact on existing deployments.
CI Status
Local harness: ✅ all 5 checks pass (frontend lint, backend lint, frontend typecheck, frontend unit tests, frontend build). GitHub CI: the PR's authoritative Build, smoke, and scan (linux/amd64 & linux/arm64), e2e, and python-analyze jobs were reported PENDING on head 2d36895 and must go green before merge (that per-arch smoke job is the only place the assembled appliance is actually booted). Additionally, the two prior human approvals were self-dismissed, so reviewDecision is REVIEW_REQUIRED — a fresh human approval is needed at merge time per GitHub policy, independent of this automated review.
UI Testing — Variant Results
✅ local: All executable code paths (59 unit tests, secret generation idempotency, probe.py, config guards, all shell syntax) pass with correct positive and negative behavior; full appliance boot is covered by CI, not reproducible in this sandbox.
✅ hosted: All reproducible test-plan items independently pass (59/59 unit tests, bash -n, negative-path present); appliance runtime itself is only exercisable via the per-arch CI smoke job, not the dev stack, so no blocking defect found with a specific file/line.
|
Final local verification passed on exact head
Login UI:
Internal tooling remains private (
|
github-project-automation
Bot
moved this from 👍🏼 Mergeable
to ✅ Done
in AutoGPT development kanban
|
🧹 Auto-undeploying: PR closed with active deployment. Cleaning up development environment for PR #13758. |
|
🧹 Preview Environment Cleaned Up All resources for PR #13758 have been removed:
Cleanup completed successfully. |
4 participants
Why / What / How
Why
AutoGPT Platform should be runnable as a complete local appliance with one
docker run, including FalkorDB-backed memory, without changing the images or deployment behavior used by hosted infrastructure.What
Adds a dedicated single-container appliance containing the frontend, backend services, PostgreSQL, a three-node Valkey cluster, RabbitMQ, FalkorDB, nginx, Supervisor, and Tini.
The appliance persists application and service state under
/data, generates installation secrets on first boot, and exposes only nginx on container port3000.How
Buildx Bake builds the existing backend
servertarget, then uses it as the base forautogpt_platform/single-container/Dockerfile. Internal services bind to loopback and Supervisor manages their lifecycle. Health checks and a fail-stop watchdog allow Docker restart policies to recover the appliance when a required service cannot recover internally.The frontend server runs as its own Unix user with a clean environment and a passwordless local PostgreSQL role limited to the Better Auth tables and required
Usercolumns. It does not receive PostgreSQL superuser credentials. Internal AppService traffic stays loopback-only inside the appliance trust boundary.The effective PR diff adds only
autogpt_platform/single-container/plus its CI workflow and smoke script. Existing backend source, frontend source, Dockerfiles, the root.dockerignore, and Compose files do not differ fromdev.Changes 🏗️
/data.linux/amd64andlinux/arm64build, runtime-smoke, vulnerability-scan, and secret-scan CI.Test plan
bash -nandgit diff --check.linux/amd64andlinux/arm64.docker run,/datareuse, generated-secret persistence, FalkorDB graph persistence, frontend database isolation, and service recovery on the exact current head.Checklist 📋
For code changes:
For configuration changes:
.env.defaultis already compatible with these changes.docker-compose.ymlis already compatible with these changes.Stack navigation
This appliance is based directly on
dev. Publication and end-user documentation are handled separately.Note
Medium Risk
Large new packaging surface with auth, secrets, and DB privilege boundaries; mitigated by extensive smoke tests and isolation checks, but it is still a new distribution path operators may run in production.
Overview
Introduces an experimental single-container AutoGPT Platform you can run with one
docker run, bundling the frontend, backend services, PostgreSQL, Valkey, RabbitMQ, FalkorDB (Graphiti memory on by default), nginx, and Supervisor. State and generated secrets live on a/datavolume; only port 3000 is exposed.The image is built via Docker Bake from the existing backend
servertarget plus a newautogpt_platform/single-container/stack (entrypoint/bootstrap, runtime config, least-privilege frontend DB role, loopback-pinned internal ports, health checks, watchdog). Operators get.env.examplefor public URL, signup policy, and LLM keys.CI adds
platform-single-container-docker.yml: unit tests for appliance helpers, per-arch image build, a long end-to-end smoke script (health, auth, hostilebackend.json, isolation, FalkorDB/nginx recovery), and Trivy critical vuln + secret scans. Hosted multi-container deployments are unchanged.Reviewed by Cursor Bugbot for commit 2d36895. Bugbot is set up for automated code reviews on this repo. Configure here.