The Archive · CLI · Concepts
Models and providers
The cloud providers and local engines the TALOS CLI works with, where keys are kept, and how a model is chosen.
Checked on CLI 0.3.3
A newer version is out (CLI 0.5.2): some details may differ.
TALOS ships no model: it works with the one you choose.
Providers
Section titled “Providers”- Cloud: OpenAI, Anthropic, Google Gemini, DeepSeek, OpenRouter, Mistral, Groq, xAI, Kimi, Qwen, Z.AI, MiniMax, Cerebras, Together, Fireworks, DeepInfra, Novita, Nebius, Hugging Face, Azure AI Foundry, Amazon Bedrock and Google Vertex AI.
- Local, with no key: Ollama, LM Studio and llama.cpp.
TALOS never falls back from a local model to a cloud one on its own. Each provider, with its endpoint and the variable that can hold its key, is on Providers.
You type a key in hidden input, TALOS tests it with a one-token request, and keeps it in your operating system’s
credential store — never in a file. A key already in your environment, such as OPENAI_API_KEY, is used only after
you agree once. For OpenRouter you can sign in in the browser instead of pasting a key. Every operation is on
talos provider.
The model
Section titled “The model”/model (or Alt+P) picks the model in the interactive screen; talos model use <provider:model> sets it from the
shell; --model overrides it for one run. talos model list shows each model’s context window, tool calling and
reasoning, and where that data comes from — the public catalog at models.dev, cached for 4 hours, unless you turn it
off. For models that support it, talos model effort sets the reasoning effort.
When something does not answer
Section titled “When something does not answer”talos provider test <provider> checks that a provider answers, without generating tokens, and says what to do.
talos doctor says what is ready and what is missing. See
Check your setup and report a problem.