ReviewGate publishes a container image for GitLab CI, private runners, and internal networks where installing from source is not desirable.
Build and smoke test a local image:
docker build -t reviewgate:local .
docker run --rm reviewgate:local --version
docker run --rm reviewgate:local doctorThe ReviewGate binary is installed at:
/usr/local/bin/reviewgate
The image starts as a non-root user and defaults to reviewgate --help.
Use the Docker image directly in merge request pipelines:
reviewgate:
stage: review
image: ghcr.io/anggaprytn/review-gate:v0.1.0-alpha.3
variables:
REVIEWGATE_LLM_PROVIDER: "ollama"
OLLAMA_BASE_URL: "http://ollama:11434"
script:
- reviewgate doctor
- reviewgate review --ci --publish
rules:
- if: '$CI_PIPELINE_SOURCE == "merge_request_event"'For early rollout, use --soft-fail so ReviewGate reports findings without failing the pipeline:
script:
- reviewgate review --ci --publish --soft-failSee ../examples/gitlab-ci-docker-reviewgate.yml for a complete example.
ollama is the recommended Docker and CI provider for private networks. Run Ollama where the CI runner can reach it and set OLLAMA_BASE_URL to that internal endpoint.
gemini_cli and codex_cli inside Docker are not recommended unless their authentication is explicitly mounted and understood. Do not mount local authentication tokens into CI containers by default.
Privacy depends on the provider mode you choose. With ollama, model inference can stay inside your machine or private network. With gemini_cli or codex_cli, the sanitized review payload is still sent to the external model service used by that CLI.
ReviewGate does not bake tokens, .env, .reviewgate/, docs/PRD.md, or local build output into the Docker image.