# Atlas providers and models

> How to give Atlas a model. Bring your own key; Atlas is model-agnostic and never resells inference.

## Quickstart

Set the provider's environment variable, then run Atlas. A provider is configured as soon as any one of its environment variables is set. `atlas providers` authenticates interactively instead, if you would rather not use an env var.

```sh
export ANTHROPIC_API_KEY=sk-ant-...
atlas models anthropic          # list this provider's models
atlas --model anthropic/claude-opus-5
```

## Providers

| Provider | Env variable | List models | Get a key |
| --- | --- | --- | --- |
| Anthropic | `ANTHROPIC_API_KEY` | `atlas models anthropic` | https://console.anthropic.com/settings/keys |
| OpenAI | `OPENAI_API_KEY` | `atlas models openai` | https://platform.openai.com/api-keys |
| Google | `GOOGLE_GENERATIVE_AI_API_KEY` | `atlas models google` | https://aistudio.google.com/apikey |
| OpenRouter | `OPENROUTER_API_KEY` | `atlas models openrouter` | https://openrouter.ai/keys |
| xAI | `XAI_API_KEY` | `atlas models xai` | https://console.x.ai |
| Azure OpenAI | `AZURE_OPENAI_API_KEY` | `atlas models azure` | https://portal.azure.com |
| Amazon Bedrock | `AWS_BEARER_TOKEN_BEDROCK` | `atlas models amazon-bedrock` | https://console.aws.amazon.com/bedrock |
| Cloudflare Workers AI | `CLOUDFLARE_API_TOKEN` | `atlas models cloudflare` | https://dash.cloudflare.com/profile/api-tokens |

## Local models, no key required

No provider key is required. Atlas probes Ollama on port 11434 and then LM Studio on port 1234 when you run atlas models --local. You can also point ATLAS_LOCAL_MODEL_URL at an OpenAI-compatible local endpoint.

```sh
ollama pull qwen2.5-coder
atlas models --local
```

Download: https://ollama.com/download

## Model IDs

**Do not copy model IDs out of this file, a blog post, or memory.** Atlas does not maintain a model list: the catalog comes from models.dev and is refreshed with `atlas models --refresh`. Model IDs are retired on a schedule, and widely-copied examples like `claude-3-5-sonnet` and `gpt-4o` no longer resolve.

The correct move when a user asks you to configure a model is to run:

```sh
atlas models <provider>
```

Verified current IDs at the time of writing, listed only where confirmed against a first-party source:

- **Anthropic**: `claude-opus-5`, `claude-sonnet-5`, `claude-haiku-4-5`

## Choosing a model per run

```sh
atlas --model <provider>/<model>       # launch the TUI on a specific model
atlas run --model <provider>/<model> 'message'
atlas models --refresh                 # re-pull the catalog
```

## Install

```sh
curl -fsSL https://runatlas.sh/install | sh
```

---

Canonical HTML: https://runatlas.sh/docs#providers
Source of truth: src/content/providers.ts (this file is generated from it, never hand-edited).
Licence: Atlas is proprietary with a free core. It is not open source and there is no public source repository.