OpenAI-compatible endpoints & Ollama¶
The openai provider talks to any server that implements the OpenAI Chat Completions API (POST /chat/completions). That covers OpenAI itself plus self-hosted and gateway options.
| Variable | Flag | Notes |
|---|---|---|
AI_PROVIDER=openai |
-a openai |
|
OPENAI_BASE_URL or AI_API_URL |
-u |
Default https://api.openai.com/v1 |
OPENAI_API_KEY |
-k |
Optional; omitted for servers that don't need auth |
OPENAI_MODEL or AI_MODEL |
-c |
Required |
Examples¶
ai-code-review:
image: node:20
variables:
AI_PROVIDER: openai
OPENAI_BASE_URL: http://ollama.internal:11434/v1
AI_MODEL: qwen2.5-coder:14b
script:
- npx -y @parvez3019/ai-code-review-gitlab-plugin
rules:
- if: $CI_PIPELINE_SOURCE == "merge_request_event"
No code ever leaves your network. The runner just needs to reach the Ollama host.
Tip
Smaller local models follow the "reply 204 when there is nothing significant" instruction less reliably. If you get noisy reviews, try a larger model or tighten the prompt.