AI Providers

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

Open full interactive viewer →

Model tiers

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

• Fast model: Used for Stage 1 (component resolution) and Stage 2 (dependency scan). These are cheaper, faster models suited for extraction tasks.
• Advanced model: Used for Stage 3 (PR analysis) and Stage 4 (model generation). These are stronger models that handle the deeper architectural reasoning.

Default models

Gemini

Tier	Default model
Fast	gemini-2.5-flash
Advanced	gemini-2.5-flash

OpenAI

Tier	Default model
Fast	gpt-4.1-mini
Advanced	gpt-4.1

Anthropic (experimental)

Tier	Default model
Fast	claude-haiku-4-5-20251001
Advanced	claude-sonnet-4-5-20250929

Caution

Anthropic support is experimental and may not produce consistent results across all codebases. Use Gemini or OpenAI for production workflows.

Overriding models

You can override the default models with environment variables:

Variable	Description
GEMINI_FAST_MODEL	Gemini model for fast tier (Stages 1–2)
GEMINI_ADVANCED_MODEL	Gemini model for advanced tier (Stages 3–4)
OPENAI_FAST_MODEL	OpenAI model for fast tier (Stages 1–2)
OPENAI_ADVANCED_MODEL	OpenAI model for advanced tier (Stages 3–4)
ANTHROPIC_FAST_MODEL	Anthropic model for fast tier (Stages 1–2)
ANTHROPIC_ADVANCED_MODEL	Anthropic model for advanced tier (Stages 3–4)

Timeout configuration

Variable	Default
GEMINI_TIMEOUT	60000 ms
OPENAI_TIMEOUT	60000 ms
ANTHROPIC_TIMEOUT	60000 ms

These control the maximum wait time for each API request. Increase them if you experience timeouts with large diffs.

Choosing a provider

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

Start with Gemini or OpenAI during evaluation.