Model Providers & Settings
Silicon Worker connects to multiple providers through a unified model gateway, so you can mix domestic and international models and set a primary plus backup models for failover. This page lists the supported provider types, the fields each requires, and how primary/backup switching works. For the setup steps, see Configure providers and Connect Your First Model.
Protocol types
Silicon Worker supports two connection protocols; most providers use the OpenAI-compatible one:
| Protocol | Notes | Auth |
|---|---|---|
| OpenAI-compatible | The most common; nearly all providers (including Chinese providers, Azure, and local models) use this | Bearer token (API key) |
| Anthropic native | The native messages API for the Claude series | x-api-key |
Supported providers
| Provider | Protocol | Typical use |
|---|---|---|
| OpenAI | OpenAI-compatible | GPT-series models |
| Anthropic | Anthropic native | Claude-series models |
| DeepSeek | OpenAI-compatible | Chinese LLM |
| Qwen / Tongyi (Alibaba) | OpenAI-compatible | Chinese LLM |
| Zhipu | OpenAI-compatible | Chinese LLM |
| SiliconFlow | OpenAI-compatible | Chinese aggregated LLMs |
| Azure OpenAI | OpenAI-compatible | Enterprise cloud deployment |
| Ollama / local models | OpenAI-compatible | Models running offline on your machine |
Fields to fill in
Whichever provider you pick, configuring a model usually only requires these fields:
| Field | Notes | Required |
|---|---|---|
| Base URL | The provider's API address. OpenAI-compatible usually points to /v1; local models point to a local address (e.g. Ollama's default local port). | Required |
| API Key | The access key from the provider's console. Local models can often leave this blank or use a placeholder. | Usually required |
| Model name | The exact model identifier to call; must match the name the provider gives. | Required |
Not sure what to put for Base URL and model name? Check your provider's official docs; examples for common providers are in Configure providers.
Primary & backup models
Silicon Worker lets you set a primary model and backup models per session:
| Setting | Effect |
|---|---|
| Primary model | The default model for the session. |
| Backup model | The model automatically switched to when the primary call fails; you can set several to fall back through in order. |
When the primary model hits a transient problem such as a network error, rate limit, or timeout, the gateway retries automatically and, if it still fails, switches to a backup model to keep the task going.