feat: Add OpenCode Go as an LLM provider

[vmhq]

Summary

Request to add OpenCode Go as a built-in LLM provider option in Craft Agents.

What is OpenCode Go?

OpenCode Go is an OpenAI-compatible API endpoint that provides access to a curated set of coding-agent models (DeepSeek, GPT, Claude, Gemini, etc.) optimized for code generation and agent orchestration tasks.

Provider Details

Field	Value
Name	OpenCode Go
Type	OpenAI-compatible
Endpoint	https://opencode.ai/zen/go/v1
API	/chat/completions (OpenAI-compatible)
Auth	API Key (Bearer token)
Website	https://opencode.ai/docs/providers/

Why this would be valuable

• Coding-agent optimized models: Models are pre-selected and benchmarked for coding agent tasks
• OpenAI-compatible: Minimal integration effort — uses the standard /v1/chat/completions endpoint
• Subscription-based: Access models through an OpenCode subscription
• Multi-model access: Access to low-cost open models through a single endpoint

How to use it manually today

Users can already configure it as a Custom provider with:

• Base URL: https://opencode.ai/zen/go/v1
• API Key: User-provided OpenCode API key
• Models: deepseek-v4-pro, kimi-k2.6, etc.

Request

Add OpenCode Go as a first-class provider option in the connection setup UI (similar to how OpenRouter, Ollama, and other third-party providers are listed), with:

1. Pre-configured base URL (https://opencode.ai/zen/go/v1)
2. Built-in model discovery/suggestions
3. Auth flow for API key input

References

• OpenCode Go Documentation
• Mastra OpenCode Go integration
• OpenCode Zen & Go (DeepWiki)

[rjulius23]

Can't you already add it as a Custom endpoint ?

[vmhq]

Yes, I was able to get it working as a Custom endpoint using https://opencode.ai/zen/go/v1 as the base URL — so it does work today as a workaround.

That said, the point of this request is first-class support, similar to how OpenRouter or Ollama are listed as built-in options. The benefits would be:

• No manual URL entry — users don't need to know the exact endpoint
• Model discovery/suggestions — surfacing available models (deepseek-v4-pro, kimi-k2.6, etc.) without having to look them up separately
• Lower friction onboarding — especially for less technical users who might not know OpenCode Go exists or how to configure a custom endpoint

Happy to keep using the custom workaround in the meantime, but native support would be a nice quality-of-life improvement! 🙏