This project supports multiple LLM providers. You only need ONE provider to use tcm, but you can configure several and switch at any time.
| Provider | Default model in tcm | Config key for API key | Environment variable | Default base URL |
|---|---|---|---|---|
| Anthropic | claude-sonnet-4-5-20250929 |
llm.api_key |
ANTHROPIC_API_KEY |
(Anthropic SDK default) |
| OpenAI | gpt-4o |
llm.openai_api_key |
OPENAI_API_KEY |
(OpenAI default) |
| DeepSeek | deepseek-v3.2 |
llm.deepseek_api_key |
DEEPSEEK_API_KEY |
https://api.deepseek.com/v1 |
| Kimi (Moonshot) | kimi-k2.5 |
llm.kimi_api_key |
MOONSHOT_API_KEY |
https://api.moonshot.cn/v1 |
| MiniMax | minimax-m2.5 |
llm.minimax_api_key |
MINIMAX_API_KEY |
https://api.minimax.chat/v1 |
| Qwen (DashScope) | qwen3-max |
llm.qwen_api_key |
DASHSCOPE_API_KEY |
https://dashscope.aliyuncs.com/compatible-mode/v1 |
| Google (Gemini) | gemini-2.5-flash (default), gemini-2.5-pro, gemini-2.5-flash-lite, gemini-2.0-flash, gemini-3.1-pro-preview |
llm.google_api_key |
GOOGLE_API_KEY |
(google-genai SDK) |
| Mistral | mistral-large-latest |
llm.mistral_api_key |
MISTRAL_API_KEY |
https://api.mistral.ai/v1 |
| Groq | llama-3.1-70b-versatile |
llm.groq_api_key |
GROQ_API_KEY |
https://api.groq.com/openai/v1 |
| Cohere | command-r-plus |
llm.cohere_api_key |
COHERE_API_KEY |
(Cohere SDK) |
Note: Together and Ollama keys can be stored via tcm keys, though they may not be active runtime providers yet.
# Show the status of all providers
tcm keys
# Set a provider key (prompts securely)
tcm keys set -p openai
tcm keys set -p kimi
tcm keys set -p deepseek
# Make a provider default for new sessions (optional)
tcm keys set -p openai --make-default
# Switch the active model (provider is auto-detected)
tcm model set gpt-4o
tcm model set kimi-k2.5
tcm will set the appropriate base URL automatically.llm.deepseek_base_url).tcm model google to fetch the live list of all available models from the API. Use the returned model IDs directly with tcm model set.gemini-3.1-pro-preview) may be subject to access restrictions or transient network issues.