Model Providers Reference

The complete Shipfox model provider catalog, including provider IDs, default models, compatible harnesses, and resolution rules.

Agent steps run against an AI provider. This page is the canonical list of every supported provider, the provider ID you use in workflow YAML, its compatible harnesses, and the default model Shipfox uses when a step doesn't name one. For a setup walkthrough, see Agent Setup.

Supported providers

Each provider is configured per workspace with its credentials. The default model is used when an agent step names the provider but not a model. A step fails when it selects a provider that is not compatible with its harness.

Provider	`provider` ID	Default model	Compatible harnesses
Anthropic	`anthropic`	`claude-opus-4-8`	`pi`, `claude`
Ant Ling	`ant-ling`	`Ring-2.6-1T`	`pi`
Azure OpenAI	`azure-openai-responses`	`gpt-5.5-pro`	`pi`
OpenAI	`openai`	`gpt-5.5-pro`	`pi`
DeepSeek	`deepseek`	`deepseek-v4-pro`	`pi`
NVIDIA	`nvidia`	`nvidia/nemotron-3-ultra-550b-a55b`	`pi`
Google AI Studio	`google`	`gemini-3.1-pro-preview`	`pi`
Mistral	`mistral`	`mistral-large-latest`	`pi`
Groq	`groq`	`openai/gpt-oss-120b`	`pi`
Cerebras	`cerebras`	`gpt-oss-120b`	`pi`
Cloudflare AI Gateway	`cloudflare-ai-gateway`	`claude-opus-4-8`	`pi`
Cloudflare Workers AI	`cloudflare-workers-ai`	`@cf/moonshotai/kimi-k2.7-code`	`pi`
xAI	`xai`	`grok-4.3`	`pi`
OpenRouter	`openrouter`	`anthropic/claude-opus-4.8`	`pi`
Vercel AI Gateway	`vercel-ai-gateway`	`anthropic/claude-opus-4.8`	`pi`
Z.ai	`zai`	`glm-5.2`	`pi`
Z.ai Coding CN	`zai-coding-cn`	`glm-5.2`	`pi`
OpenCode	`opencode`	`claude-opus-4-8`	`pi`
OpenCode Go	`opencode-go`	`kimi-k2.7-code`	`pi`
Hugging Face	`huggingface`	`deepseek-ai/DeepSeek-V4-Pro`	`pi`
Fireworks	`fireworks`	`accounts/fireworks/models/deepseek-v4-pro`	`pi`
Together AI	`together`	`deepseek-ai/DeepSeek-V4-Pro`	`pi`
Kimi Coding	`kimi-coding`	`k2p7`	`pi`
MiniMax	`minimax`	`MiniMax-M3`	`pi`
MiniMax CN	`minimax-cn`	`MiniMax-M3`	`pi`
Moonshot AI	`moonshotai`	`kimi-k2.7-code`	`pi`
Moonshot AI CN	`moonshotai-cn`	`kimi-k2.7-code`	`pi`
Xiaomi	`xiaomi`	`mimo-v2.5-pro`	`pi`
Xiaomi Token Plan CN	`xiaomi-token-plan-cn`	`mimo-v2.5-pro`	`pi`
Xiaomi Token Plan AMS	`xiaomi-token-plan-ams`	`mimo-v2.5-pro`	`pi`
Xiaomi Token Plan SGP	`xiaomi-token-plan-sgp`	`mimo-v2.5-pro`	`pi`

Custom model providers are workspace-specific, so they do not appear in this table. They are compatible with the pi harness. Use Add a custom provider to connect one.

Most providers need only an API key. Azure OpenAI also requires an endpoint. Cloudflare providers require an account ID (and a gateway ID for the AI Gateway). The configuration form shows the fields each provider needs.

Selecting a model

The model you set on an agent step is passed to the provider and validated against that provider's available models: an unrecognized model ID fails the step with a clear error. Each provider's full model list comes from the provider's own catalog. The default model above is what Shipfox picks when you omit model.

How configuration works

Providers are configured per workspace. For each provider you configure, you supply its credentials and can optionally set a default model and mark one provider as the workspace default. Credentials are stored securely. Shipfox stores their encrypted values and never returns the raw secrets through settings or management APIs. A runner receives the decrypted credential bundle only for its currently leased agent step. Configuration takes effect immediately. No redeploy is needed.

Resolution and defaults

When an agent step runs, Shipfox resolves the harness, provider, model, and thinking level in this order:

Harness: the step's harness, else the workspace default harness, else pi.
Provider: the step's provider, else a compatible workspace default provider, else anthropic.
Model: the step's model, else the configured default model for the resolved provider, else that provider's catalog default (the table above).
Thinking: the step's thinking, else the provider's configured default, else xhigh.

So an agent step with just a prompt runs on the workspace default harness, provider, and model. Naming a provider but no model uses that provider's default model.

Unsupported providers

Amazon Bedrock, Google Vertex AI, OpenAI Codex, and GitHub Copilot appear in the catalog but are not yet supported: they require cloud IAM credentials or OAuth flows that API-key configuration can't provide. Support is planned for a future release.

Agent setup

Configure a harness, provider credentials, and workspace defaults.

Agent steps

Select harnesses, models, thinking levels, and providers on an agent step.

Was this page helpful?

Edit this page on GitHub