Quick Start
Use the quick start to boot the operator server, verify readiness, then trigger one exact PR or MR review. Webhook automation is optional; the default operating model is manual first.
Before you start
7review needs one SCM provider, one model path, and the production sidecars used for context work.
| Requirement | Purpose |
|---|---|
| GitHub or GitLab token | Fetch PR/MR metadata, diff, files, and publish review output |
| Model provider configuration | Run the reasoner and any configured review roles |
HEADROOM_URL | Reduce and fit source context before model review |
MEMPALACE_URL | Recall and store approved repository memory |
REVIEW_API_TOKEN | Protect operator tools such as request_review |
Run setup first so the local .env template contains the expected keys:
go run ./cmd/7review setup
Start the server
- Local
- Docker
set -a
. ./.env
set +a
go run ./cmd/7review
In another terminal, check that the HTTP server and operator surface are alive:
go run ./cmd/7review status --server http://localhost:8080
make docker-up
make docker-status
Use the raw readiness target when you only need the /ready result:
make docker-ready
Trigger the first review
Manual review is the clearest first run because it names one exact change and uses the same worker queue as webhook work.
- GitHub
- GitLab
7review review github \
--repo owner/repo \
--pr 7 \
--server http://localhost:8080
The run ID is owner/repo!7.
7review review gitlab \
--project 25 \
--mr 19 \
--server http://localhost:8080
The run ID is 25!19.
Inspect progress
7review sessions --server http://localhost:8080
7review session owner/repo!7 --server http://localhost:8080
7review history owner/repo!7 --server http://localhost:8080
Use chat when you want to inspect or continue from a specific run context:
7review chat owner/repo!7 --server http://localhost:8080
What success looks like
After the run starts, 7review should:
- enrich the PR or MR from GitHub or GitLab
- collect the structured diff and selected source context
- call Headroom and MemPalace
- route review through the configured model roles
- validate findings against the current diff
- prepare draft review output
- stop before final publication until human approval is recorded
If the request is rejected, check whether the same run ID is already queued or running, whether the queue is full, or whether the provider identifiers are wrong.