Supported Models

Supported Models

How Refact discovers models and applies capability metadata.

Refact does not maintain a static public list of every usable model. Model availability changes frequently, so Refact discovers models from configured providers and combines them with local capability metadata in the engine.

Supported Provider Registry

Refact currently includes provider support for:

Provider	Model source
Anthropic	Anthropic Messages API and provider model metadata.
OpenAI	OpenAI Chat Completions.
OpenAI Responses	OpenAI Responses.
OpenAI Codex	Codex-oriented OpenAI Responses models.
OpenRouter	OpenRouter catalog and OpenAI-compatible endpoints.
Ollama	Local runtime model list and custom settings.
LM Studio	Local runtime model list and custom settings.
vLLM	Self-hosted OpenAI-compatible endpoints and custom settings.
Groq	Groq OpenAI-compatible endpoints.
DeepSeek	DeepSeek OpenAI-compatible endpoints.
Doubao	Volcengine/Doubao endpoints.
xAI	xAI Chat Completions and Responses endpoints.
Gemini	Google Gemini OpenAI-compatible endpoints.
Qwen	DashScope-compatible endpoints.
Kimi	Moonshot/Kimi endpoints.
Zhipu	Zhipu/Z.ai endpoints.
MiniMax	MiniMax Anthropic-compatible endpoint.
GitHub Copilot	GitHub Copilot chat endpoint.
Custom	User-defined OpenAI-compatible endpoints.
Claude Code	Claude Code/Anthropic-compatible provider flow.

How Refact Decides What A Model Can Do

Refact combines provider model discovery with capability metadata. That metadata can describe:

• Context window size.
• Tool and agent support.
• Vision or multimodal support.
• Reasoning and thinking options.
• Tokenizer and scratchpad behavior.
• Completion support.
• Embedding support.
• Provider-specific request format and endpoint details.

The UI uses these capabilities when you choose default chat, reasoning, completion, and embedding models.

Dynamic Discovery Instead Of Fixed Claims

Some providers expose model lists through an API. Others use bundled provider templates, custom entries, or local runtime responses. A model can be usable even if it is not named in this documentation, provided the provider accepts it and Refact has enough capability information to route requests safely.

If a provider adds a new model before Refact has bundled metadata for it, configure a custom model entry in the provider settings. Include the context size, tokenizer, tool support, completion support, and other settings relevant to your workflow.

Local Models

For Ollama, LM Studio, and vLLM:

1. Install or download the model in the runtime.
2. Start the runtime and confirm the endpoint is reachable.
3. Add the runtime provider in Refact.
4. Enable the model and choose it in Default Models.

Local runtimes can be used for chat, agents, completion, or embeddings when the runtime and model support the needed capability.

Pricing And Usage

Refact does not sell model access. Any cost shown in the UI is a local estimate based on provider pricing metadata where available. Actual billing, quotas, model access, rate limits, and retention policies are handled by the provider or runtime you configure.