Provider matrix

When to read this: Configure LLM backends, auth, or default model capabilities.

Source: crates/chatty-core/src/settings/models/providers_store.rs, crates/chatty-core/src/factories/agent_factory/provider_builder.rs.

Provider types

Provider	Display name	Auth	Required config	Default images	Default PDF	Temperature default
`openrouter`	OpenRouter	API key	`api_key`; optional `base_url`	yes	yes	per `ModelConfig`
`ollama`	Ollama	none	optional `base_url` (default `http://localhost:11434`)	no	no	always applied
`azure_openai`	Azure OpenAI	API key or Entra ID	`base_url` (resource endpoint) + credentials	yes	no	per `ModelConfig`

Gateway to 200+ upstream models (Anthropic, OpenAI, Google, Mistral, Meta, …).
Persisted ProviderType JSON is open_router (serde snake_case). Legacy values open_ai, open_a_i, anthropic, gemini, mistral deserialize as OpenRouter. The string openrouter is not accepted.
configured_providers() requires a non-empty api_key.
Rig client: rig_core::providers::openrouter::Client.
MCP tools are sanitized for OpenAI-compatible schemas before attachment.

Local or remote Ollama instance; no API key required.
Included in configured_providers() even without api_key.
Default capabilities are (false, false); vision/PDF support is detected per model via /api/show (sync_service.rs) and stored in ModelConfig.
Rig client: rig_core::providers::ollama::Client.

Auth via extra_config.auth_method: api_key (default) or entra_id.
configured_providers() requires non-empty base_url and (api_key or Entra ID).
Endpoint URLs are normalized (trailing slashes stripped, /openai/deployments/... paths truncated).
Per-model api_version in ModelConfig.extra_params; default from AZURE_DEFAULT_API_VERSION.
Entra ID tokens are cached in AzureTokenCache when available.
Rig client: rig_core::providers::azure::Client.

Field	Purpose
`supports_images`	Allow image attachments in chat
`supports_pdf`	Allow PDF attachments
`supports_temperature`	Show temperature slider; skipped for reasoning models

New models inherit ProviderType::default_capabilities(). Ollama overrides per model after sync.

Flag	Maps to	Notes
`--ollama [URL]`	Ollama provider	Auto-discovers models from `/api/tags`
`--openai-compat-url URL`	OpenRouter provider type	vLLM, llama.cpp, LM Studio, etc.; optional `--api-key`

See CLI flags for full chatty-tui options.