LLM Providers and Configuration

LLM Providers & Configuration 🤖

Canary is designed to be LLM-agnostic. You can switch between different AI backends without changing the core test generation logic.

The Factory Pattern

The ProviderFactory selects the appropriate backend based on the CANARY_LLM_PROVIDER environment variable.

Available Providers

Provider	Description	Required Configuration
anthropic (Default)	Uses Claude Sonnet for high-quality test generation.	ANTHROPIC_API_KEY
gemini	Google Gemini Flash for fast, low-cost generation.	GEMINI_API_KEY
openai	Uses GPT-4o-mini, retained as an alternative.	OPENAI_API_KEY
mock	Returns static test templates. Used by tests and CI.	None

How to Configure

Switching Providers

To switch to a non-default provider:

export CANARY_LLM_PROVIDER='gemini'

The default is anthropic (Claude) — no env-var needed to use it.

Setting API Keys

Canary uses lazy initialization: it only loads the provider SDK and checks for an API key when generation actually runs. The CLI can run version or --recommend-only commands with no key set, and missing optional SDKs (e.g. google-generativeai) only break the provider that needs them, not the rest of the system.

export ANTHROPIC_API_KEY='your-key-here'

Creating a New Provider

To add a new LLM backend:

1. Create a new file in agent/llm/providers/.
2. Inherit from BaseProvider and implement generate(messages).
3. Register the new provider in _PROVIDER_REGISTRY in agent/llm/factory.py as (module_path, class_name).