03 / Bring your own LLM

Your model. Our review. No middleman.

Most AI review tools lock you into their inference stack. CodePatrol does the opposite: point us at any OpenAI-compatible endpoint — Ollama on a workstation, vLLM in your cluster, Azure OpenAI in West Europe, Mistral in Paris — and we handle the rest.

codepatrol.config.yaml

# Bring your own inference
provider:
  type: openai-compatible
  endpoint: https://llm.internal.eu
  model: qwen3-coder-30b
  api_key: ${VAULT_LLM_KEY}
review:
  focus:
    - security
    - performance
  skip_paths:
    - /vendor/**
✓ connected · 30b model · 47k context

Why this matters now

The inference layer is becoming a commodity. Tools that lock you out of it are not.

Qwen, Mistral, Llama, DeepSeek — open-weight models now match GPT-class performance on code tasks. Running them on your hardware is a tractable engineering problem, not a frontier one.

Yet every major AI review product still bills you for inference they don't disclose, runs it where you can't see, and locks you into their model choices. We think that's the wrong abstraction.

What we connect to

Anything that speaks OpenAI-compatible.

Self-hosted

Ollama
vLLM
LM Studio
Text Generation Inference

European cloud

Azure OpenAI (Sweden, Netherlands)
Mistral La Plateforme
OVHcloud AI Endpoints
Scaleway Inference

Open weight

Qwen3 Coder
DeepSeek
Llama 3
Mistral Codestral

Frontier (optional)

Anthropic Claude
OpenAI
Google Gemini
—

The only requirement is OpenAI-compatible chat completions. If your stack can talk to it, CodePatrol can use it.

How it fits together

We're the review brain. You own the eyes.

Install the GitHub app

Or self-host the connector for GitLab / Forgejo / Gitea / Bitbucket.

Point it at your LLM

OpenAI-compatible endpoint. URL, key, model name. No vendor approval needed.

Review runs on every PR

We handle prompt engineering, chunking, deduplication. Inference cost is yours to track.

Pricing

Flat monthly. No per-seat. No per-review.

Because we don't pay for your inference, we don't need to charge per call. Beta pricing will start at €99/month per repository, regardless of team size or PR volume.

Final pricing locks in for beta users for the first 12 months.

Technical preview · Summer 2026