install.sh now asks one question on fresh installs — the URL the app is
opened on. Accepts https://primora.example.com (TLS at the user's proxy),
primora.example.com, or 192.168.1.50:8085; the port becomes NGINX_PORT.
The answer rewrites every public URL var and COOKIE_DOMAIN for real DNS
names. PUBLIC_URL env keeps it non-interactive; legacy DOMAIN/NGINX_PORT
still work; reruns over an existing .env skip the question and display
the URL from the file.
Also trust both URL schemes on the serving host in trustedOrigins —
external TLS termination makes nginx report http while browsers arrive
over https.
Generated with [Devin](https://devin.ai)
Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>
A "Set up with an AI agent" dialog that mints a write-scoped API key and
renders a copyable prompt handing a coding agent everything needed to wire
a repository into a Primora project: telemetry ingest for any language,
db-connections for tracked databases, MCP/CLI agent access, and
verification steps. Opens at the end of new-project onboarding; also
reachable from the project dashboard and Projects page settings.
Authored By: TDvorak <[email protected]>
Generated with [Devin](https://devin.ai)
Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>
- POST /projects/:id/events accepts client-authored events restricted to
the custom.* namespace — system types can't be spoofed. They flow
through the same publishEvent fan-out: realtime, webhooks, and
event_pattern functions all fire.
- Presence: subscribe/unsubscribe bumps a per-project count and
broadcasts presence.update {online} to the realtime stream only —
deliberately not through publishEvent so webhooks and "*" functions
don't fire on connect noise. GET …/realtime/presence reads the count.
- @primora/client gains publish() and presence(); CLI gains events:send.
Verified live: presence 0->1 on subscribe, presence.update frame
delivered, custom event delivered to SSE + fired a custom.* function,
spoofed document.created rejected 400.
Generated with [Devin](https://devin.ai)
Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>
POST /buckets/:id/object-presigns mints SigV4 query-signed GET/PUT
URLs so clients move bytes directly against S3 — large transfers skip
the backend entirely.
- storage.Presigner optional interface; S3Store implements, local
driver returns a clear error (no external surface to sign).
- TTL 60s–1h (default 15m); download requires read role + existing
object row (no key probing); upload requires write role.
- S3_PUBLIC_ENDPOINT rewrites the client-facing host for internal
S3_ENDPOINTs — signatures bind the host, URLs are valid only where
minted.
- CLI: primora objects:presign <bucket> <key> [--upload] [--ttl].
Verified live against cloudserver: GET fetched the object with no
auth headers; PUT pushed 1 MB direct to S3; unsigned list → 403;
untracked-key download presign → 404.
Generated with [Devin](https://devin.ai)
Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Migration 00016: function_id on scheduled_jobs/inbound_hooks,
event_pattern on functions, 'event' in the run-trigger enum.
- Jobs deliver to a function instead of HTTP when function_id is set;
the run's status_code mirrors the process exit code. Clearing the
field returns the job to URL delivery.
- Inbound hooks gain mode "function" — received body becomes stdin
payload, acked immediately, run recorded async.
- event_pattern (document.*, *, exact type) fires the function detached
on matching publishEvent — never blocks realtime/webhook fan-out.
- One executeFunction path serves manual/schedule/hook/event; run rows
record the trigger source.
- Backend image now ships bun 1.3.14 so Functions work under compose.
Dashboard: job target picker, hook function mode, event-pattern field.
Verified live: job run → success (status_code 0), hook POST → run,
document.created → event run with correct payload.
Generated with [Devin](https://devin.ai)
Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>
README down to the Containr skeleton: header, three-screenshot row
(rest behind <details>), features, architecture, quick start, links.
Dropped tables and sections already covered by QUICK_START.md.
Native dev commands and repo layout moved to CONTRIBUTING.md;
CLI/MCP usage moved to new per-app READMEs.
Authored By: TDvorak <[email protected]>
Generated with [Devin](https://devin.ai)
Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>
- docker-build job is now a per-app matrix with GHA layer cache, a
container contents smoke check, and a non-blocking Trivy scan.
- New docker-publish job pushes edge + sha tags to GHCR on master.
- Fixed release.yml tagging ghcr.io/<Owner>/... with a mixed-case
owner — GHCR requires lowercase, so tagged releases would have
failed at push time.
- compose files: image: keys name local builds after the GHCR repos.
- install.sh: curl|bash installer — fetches compose files, generates
secrets, pulls prebuilt images (docker pull + up --no-build, since
compose pull skips buildable services), falls back to a source
tarball build when packages aren't public.
- README: install section covering the one-liner, clone, and prebuilt
paths plus common compose commands.
Verified locally: all three images build, full dev-overlay stack
passes scripts/smoke-e2e.sh, and install.sh produces a healthy
stack in a standalone directory.
Authored By: TDvorak <[email protected]>
Generated with [Devin](https://devin.ai)
Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>
