LLM Providers

Git-Iris supports three LLM providers: OpenAI, Anthropic, and Google.

Provider Overview

Provider	Default Model	Fast Model	Context Window	API Key Env
OpenAI	gpt-5.4	gpt-5.4-mini	128K	OPENAI_API_KEY
Anthropic	claude-opus-4-6	claude-haiku-4-5-20251001	200K	ANTHROPIC_API_KEY
Google	gemini-3-pro-preview	gemini-2.5-flash	1M	GOOGLE_API_KEY

Configuration Format

Each provider has its own section under [providers]:

[providers.PROVIDER_NAME]
api_key = "YOUR_API_KEY"
model = "model-name"           # Optional: primary model
fast_model = "fast-model-name" # Optional: for status updates
token_limit = 8000             # Optional: custom limit

OpenAI Configuration

[providers.openai]
api_key = "sk-..."
model = "gpt-5.4"
fast_model = "gpt-5.4-mini"
token_limit = 128000

CLI Setup

git-iris config --provider openai --api-key YOUR_API_KEY
git-iris config --provider openai --model gpt-5.4

Environment Variable

export OPENAI_API_KEY="sk-..."

Anthropic Configuration

[providers.anthropic]
api_key = "sk-ant-..."
model = "claude-opus-4-6"
fast_model = "claude-haiku-4-5-20251001"
token_limit = 200000

CLI Setup

git-iris config --provider anthropic --api-key YOUR_API_KEY
git-iris config --provider anthropic --model claude-opus-4-6

Environment Variable

export ANTHROPIC_API_KEY="sk-ant-..."

Legacy Aliases

The provider names claude and gemini are still supported as aliases for anthropic and google.

Google Configuration

[providers.google]
api_key = "your-google-api-key"
model = "gemini-3-pro-preview"
fast_model = "gemini-2.5-flash"
token_limit = 1000000

CLI Setup

git-iris config --provider google --api-key YOUR_API_KEY
git-iris config --provider google --model gemini-3-pro-preview

Environment Variable

export GOOGLE_API_KEY="..."

Switching Providers

Set Default Provider

git-iris config --provider anthropic

Override Per-Command

git-iris gen --provider openai
git-iris review --provider google

Additional Parameters

Provider-specific parameters can be set using --param:

git-iris config --provider openai --param reasoning='{"effort":"medium"}'
git-iris config --provider openai --param text='{"verbosity":"low"}'

Git-Iris parses valid JSON values here, so nested provider options work without extra config files. For OpenAI reasoning models, use --token-limit to control output-token budgets; Git-Iris maps that to the provider's current completion-token setting automatically.

OpenAI GPT-5 defaults are already tuned by workflow:

• Main agent generations use reasoning = {"effort":"medium"}
• Subagents and parallel_analyze use reasoning = {"effort":"low"}
• Fast status messages use reasoning = {"effort":"none"}

Set reasoning yourself only when you want to override those defaults for every OpenAI request using that provider config.

In TOML:

[providers.openai]
api_key = "sk-..."

  [providers.openai.additional_params]
  reasoning = '{"effort":"medium"}'
  text = '{"verbosity":"low"}'

Token Limits

Each provider has a default context window. You can override this:

git-iris config --provider anthropic --token-limit 100000

This is useful for:

• Cost control (smaller limit = fewer tokens)
• Faster responses
• Testing with limited context

API Key Priority

API keys are loaded in this order:

1. Config file (~/.config/git-iris/config.toml)
2. Environment variable (OPENAI_API_KEY, etc.)

Project configs (.irisconfig) never contain API keys for security.

Verification

Check your provider configuration:

# View current configuration
cat ~/.config/git-iris/config.toml

# Test with a command
git-iris gen --print

If authentication fails, Git-Iris will tell you which environment variable to set.

Security Best Practices

• Never commit config.toml with API keys
• Use environment variables in CI/CD
• Restrict file permissions:

  chmod 600 ~/.config/git-iris/config.toml

• Rotate API keys periodically
• Use separate keys for different projects if needed