Sophon Docs
Models & Providers

Supported ProvidersNEW

Every model provider Sophon supports, grouped by type, with provider ids and authentication.

Sophon is bring-your-own-key: it ships with no pre-configured providers. You add the ones you want, and each provider's credentials are encrypted at rest in models.json. The tables below list every provider Sophon supports.

The id in each row is the value you set as the provider type — both in models.json (the providerType field) and in the interactive sophon models add wizard. Use these exact ids.

The model lists each provider exposes — display names, capabilities, and per-token pricing in pickers and Settings — are backed by Sophon's bundled Model Catalog, which you can extend or override with your own entries.

Auth types come from the ProviderAuthType enum: ApiKey (Bearer token), CliSync (read tokens from an external CLI's credential store), OAuth (Sophon runs OAuth 2.1 + PKCE), and None (local, no credential).

Cloud LLM providers

API key unless the Notes column says otherwise.

ProvideridAuthNotes
Anthropic (Claude)anthropicApiKey
OpenAI (GPT)openaiApiKey
Google GeminigeminiApiKey
Azure OpenAIazure-openaiApiKeyAlso needs a resource endpoint and a deployment name; optional api-version (defaults to a recent preview).
AWS BedrockbedrockApiKeyAWS access key id + secret access key + region (default us-east-1); signed with SigV4. Converse API.
Cloudflare Workers AIcloudflareApiKeyAccount id + API token. Account id is embedded in the base URL.
DeepSeekdeepseekApiKey
Fireworks AIfireworksApiKey
GroqgroqApiKey
Hugging Face InferencehuggingfaceApiKey
MiniMaxminimaxApiKey
Mistral AImistralApiKey
Moonshot AI (Kimi)moonshotApiKey
NVIDIA NIMnvidia-nimApiKey
OpenRouteropenrouterApiKeyAggregator — reaches 100+ models behind one key.
PerplexityperplexityApiKey
Qwen (Alibaba Cloud)qwenApiKey
StepFunstepfunApiKey
Together AItogetherApiKey
xAI (Grok)xaiApiKey
Zhipu AI (GLM)zhipuApiKey
Sophon Forgesophon-forgeApiKeySophon's hosted, OpenAI-compatible inference service — see Sophon Forge.

Local / self-hosted

No auth — set a base URL (a sensible default is assumed if you omit it).

ProvideridAuthNotes
OllamaollamaNoneBase URL defaults to http://localhost:11434. Tool calling depends on the model.
LM StudiolmstudioNonePoint at your local LM Studio server's OpenAI-compatible base URL.

OpenAI-compatible

For any endpoint that speaks the OpenAI Chat Completions format.

ProvideridAuthNotes
Custom (OpenAI-compatible)custom-openaiApiKeyRequires a base endpoint; API key is optional. Works with LiteLLM, vLLM, proxies, and self-hosted gateways.

Subscription / CLI sign-in

These reuse an existing subscription instead of a metered API key. Per-token cost is reported as $0 because billing is flat-rate. See Subscription & CLI Sign-In for the sign-in flow.

ProvideridAuthNotes
Anthropic via Claude Pro/Maxanthropic-subscriptionCliSyncReads Claude Code CLI credentials. Sign in with claude auth login. Flat-rate.
Claude Code engineclaude-codeCliSyncDrives the Claude Code engine using synced CLI credentials. Flat-rate.
OpenAI Codex (ChatGPT)openai-codexOAuthCodex CLI / OAuth against a ChatGPT subscription. Flat-rate.
GitHub Copilotgithub-copilotOAuthDevice-flow sign-in; exchanges for a Copilot token. Flat-rate.

models.json example

A single configured provider looks like this. sophon models add writes the same structure for you (and encrypts the key):

{
  "providers": [
    {
      "id": "anthropic-primary",
      "name": "Anthropic",
      "providerType": "anthropic",
      "apiKey": "sk-ant-...",
      "priority": 1,
      "isDefault": true,
      "models": [
        { "id": "claude-sonnet-4-5", "isDefault": true }
      ]
    }
  ]
}

priority orders providers for routing and failover; lower numbers are tried first. Combine multiple providers, set per-provider budgets, and control fallback in Routing, Failover & Budgets.

Reaching a provider that isn't listed

If a model isn't covered by a native provider above, two ids fill the gap:

  • openrouter — one API key fronts 100+ hosted models across many vendors.
  • custom-openai — point at any OpenAI-compatible base URL (self-hosted, a proxy, or a new vendor) with an optional key.