Proval

Set LLM

Add a model provider in Proval with Base URL, API key, and model ID.

Proval sends review context to the LLM endpoint you configure. Your code and review context go only to the Base URL you set. Proval does not host or proxy models.

This guide covers how to fill in the model provider form in the dashboard. For ready-made endpoint values, see the example guides:

Prerequisites

  • Proval is running (Quick Start)
  • You can open the dashboard at http://<your-server>:7900

Choose an API provider

In Model Provider → Add model provider, pick the provider that matches your endpoint:

Provider in Proval Protocol Use when
OpenAI Chat Completions API OpenRouter, llama.cpp, Ollama, vLLM, or any OpenAI-compatible server
Anthropic Messages API Anthropic directly, or a proxy that speaks the Messages API

Most self-hosted and aggregator setups use OpenAI (Chat Completions). Switch to Anthropic only when your endpoint expects the Messages API.

Fill in the model provider form

  1. Open the dashboard → Model Provider → Add model provider
  2. Complete each field:
Field What to enter
Display Name A label for your team, e.g. Main Reviewer
API Provider OpenAI or Anthropic (see above)
Model ID The model name your API expects, e.g. anthropic/claude-sonnet-4.6 or llama-3.1-8b-instruct
Base URL Root URL of the API, including /v1 when your server uses it, e.g. https://openrouter.ai/api/v1 or http://<llm-host>:8080/v1
API Key Key required by that endpoint (sk-or-..., sk-ant-..., or a placeholder if the server has no auth)
Proval Create Model Provider form with all fields filled in
Example: OpenAI provider with an OpenRouter base URL

Base URL + API key are the two values that define where requests go and how they authenticate. Get them from your provider. OpenRouter and llama.cpp walk through common cases.

Test and save

  1. Click Test Connection. You should see Connection successful
  2. Click Create
Proval connection test success dialog
Successful connection test before saving

If the test fails, double-check Base URL, Model ID, and API Key against your provider's docs. Wrong provider type (OpenAI vs Anthropic) is a common mistake.

What's next

Connect your Git host and link a repository: