Skip to content

AI Providers

Erode supports three AI providers: Gemini (default), OpenAI, and Anthropic (experimental). Set the provider with the ERODE_AI_PROVIDER environment variable.

Open full interactive viewer →

Each provider uses two model tiers to balance cost and quality:

  • Fast model: Used for Stage 1 (component resolution), Stage 2 (dependency scan), and model updates. These are cheaper, faster models suited for extraction tasks.
  • Advanced model: Used for Stage 3 (PR analysis). These are stronger models that handle the deeper architectural reasoning.
TierDefault model
Fastgemini-2.5-flash
Advancedgemini-2.5-pro
TierDefault model
Fastgpt-4.1-mini
Advancedgpt-4.1
TierDefault model
Fastclaude-haiku-4-5-20251001
Advancedclaude-sonnet-4-5-20250929

Override the default models per provider with environment variables. See Configuration: Provider model overrides for the full list.

Each provider has a configurable request timeout (default: 60 seconds). Increase them if you experience timeouts with large diffs. See Configuration: Timeouts for all timeout variables.

Gemini is the default provider and is cheaper per request. OpenAI offers strong analysis quality with broad model availability. Anthropic support is experimental.

Start with Gemini or OpenAI during evaluation.