notagent
Install

Models

Providers and local endpoints

45 built-in providers, ambient API keys, and first-class support for Ollama, LM Studio and any OpenAI-compatible server.

Built-in providers

45 providers ship in the binary, among them Anthropic, OpenAI, OpenAI Codex, Google, Google Vertex, Amazon Bedrock, Azure OpenAI, GitHub Copilot, Mistral, DeepSeek, xAI, Groq, Cerebras, Fireworks, Together, Baseten, NVIDIA, Hugging Face, OpenRouter, the Vercel and Cloudflare AI gateways, Cloudflare Workers AI, Moonshot AI, Kimi Coding, Qwen, MiniMax, Xiaomi, Z.ai, Ant Ling, OpenCode and Cline. Regional variants and token-plan endpoints are separate providers.

/login stores an API key or runs an OAuth flow. Providers are also picked up from ambient environment variables (ANTHROPIC_API_KEY, OPENAI_API_KEY, and so on) without a login.

Local and custom endpoints

Three providers take their model list from a server rather than from a built-in catalog, so they offer exactly what that server holds:

  • Ollama: /login ollama, pointed at http://127.0.0.1:11434/v1.
  • LM Studio: /login lmstudio, pointed at http://127.0.0.1:1234/v1.
  • Custom (OpenAI-compatible): /login custom asks for a base URL, which is stored with the credential, so /logout custom removes the endpoint along with the key. A URL may be typed without a scheme or without /v1; both are filled in.

The login asks for an API key in every case. On a loopback address the question can be answered with an empty line, since a local server authorizes everything until its own authentication is switched on. Both runtimes offer that, and a key typed here is sent as a bearer token from then on.

None of them appear in the model picker until they are activated, and /logout deactivates them again. The model list comes from the server, so it holds exactly what was pulled or loaded; an address nothing answers on is reported rather than passed over. Where a runtime reports a model’s context window and whether it reasons, those are used, and the effort levels offered are the ones that runtime accepts.

Offline use

The binary contains a built-in model catalog, so it remains usable offline. Catalog updates for authenticated providers are fetched from notagent.dev and cached in ~/.notagent/agent/models-store.json. See Custom providers for forcing a refresh.