Providers and Models

Athas can use hosted providers, local models, and custom OpenAI-compatible endpoints.

Where Providers Are Used

Provider settings are split by feature:

• Athas Agent uses the provider and model selected in the main AI section
• Autocomplete uses the provider and model selected in the Autocomplete section
• External Agents use their own CLI authentication outside Athas

This lets you use one model for chat and another model for inline completions.

Built-in Providers

Custom OpenAI-compatible Providers

Use Custom when a provider exposes an OpenAI-compatible chat completions API.

For Athas Agent:

1. Go to Settings > AI
2. Select Custom as the provider
3. Enter the model ID exactly as your provider expects it
4. Enter the provider base URL
5. Save an API key if the endpoint requires a bearer token

For autocomplete:

1. Go to Settings > AI > Autocomplete
2. Select Custom as the autocomplete provider
3. Enter the model ID, base URL, and optional API key

The base URL can point to either the API root or the chat completions endpoint. For example, https://api.example.com/v1 and https://api.example.com/v1/chat/completions are both valid.

Autocomplete Model Lists

When autocomplete uses OpenRouter, Athas loads the model list from the live Athas/OpenRouter model endpoint. If the list cannot be loaded, Athas leaves the selector empty and lets you retry.

Athas does not show hardcoded fallback models as selectable models.

Model Parameters

• Temperature: lower values are more deterministic, higher values are more creative
• Max Tokens: limits response length