Skip to main content

Configure LLM

Cognitive Modules (Node CLI) supports multiple LLM providers. The CLI auto-selects a provider based on which API key is present (stable providers only), or you can force one with --provider.

Notes:

  • If multiple provider API keys are set, the CLI will auto-select the first provider it finds. Prefer passing --provider ... explicitly to avoid surprises.
  • Auto-select uses a fixed priority order: OpenAI -> Anthropic -> Gemini -> MiniMax -> DeepSeek -> Qwen. If your account does not support the default model for the selected provider, pass --model ... or choose a different provider.

Supported Providers​

Stable support surface (guaranteed by docs/CI/release gates):

Provider (Stable)Environment Variable
OpenAI (ChatGPT)OPENAI_API_KEY
Anthropic (Claude)ANTHROPIC_API_KEY
GeminiGEMINI_API_KEY
MiniMaxMINIMAX_API_KEY
DeepSeekDEEPSEEK_API_KEY
Qwen (DashScope)DASHSCOPE_API_KEY or QWEN_API_KEY

Experimental/community providers are still available, but require explicit --provider and are not part of the stable promise:

Provider (Experimental/Community)Environment Variable
Moonshot (Kimi)MOONSHOT_API_KEY
Ollama (local)OLLAMA_HOST

Example: OpenAI​

export OPENAI_API_KEY=sk-xxx
npx cogn@2.2.16 run code-reviewer --args "your code" --provider openai --model gpt-4o

Example: Anthropic​

export ANTHROPIC_API_KEY=sk-ant-xxx
npx cogn@2.2.16 run code-reviewer --args "your code" --provider anthropic --model claude-sonnet-4.5

Example: Ollama (Local) (Community)​

# Install and start Ollama separately
export OLLAMA_HOST=http://localhost:11434
npx cogn@2.2.16 run code-reviewer --args "your code" --provider ollama --model llama3.1

Model Override​

  • CLI: --model <name>
  • Env: COG_MODEL (global override)

Check Configuration​

npx cogn@2.2.16 doctor