Feature Request: Support OpenAI-Compatible Providers for Claude Code

[luis15pt]

Summary

Add support for any OpenAI-compatible API provider as a selectable option in the Connect Your AI Agents setup screen. Users would click Login, enter their API key and endpoint, select a model, and Claude Code sessions route through that provider — no external proxy or manual configuration needed.

This includes self-hosted models — any local or on-premises inference server that exposes the standard /v1/chat/completions endpoint works out of the box.

Problem

Claude Code only speaks Anthropic API format (/v1/messages). This limits HolyClaude users to Anthropic (direct) or Ollama (which added Anthropic compatibility). The vast majority of third-party inference providers use the OpenAI API format (/v1/chat/completions), which is the de facto industry standard.

Users who want to use these providers today must manually set up an external translation proxy (e.g. LiteLLM), which defeats the purpose of HolyClaude's "just works" philosophy.

Proposal

HolyClaude handles the API translation internally. When a user selects an OpenAI-compatible provider, a built-in lightweight proxy converts Anthropic ↔ OpenAI format behind the scenes. The user never sees or manages this.

Setup screen

☐ Claude Code           — Not authenticated            [Login]
☐ Cursor                — Command timeout               [Login]
☐ OpenAI Codex          — Codex not configured           [Login]
☐ Gemini                — Gemini CLI not configured      [Login]
☐ OpenAI-Compatible     — Not configured                 [Login]   ← NEW

Login flow

1. Click Login next to OpenAI-Compatible
2. Enter API base URL (e.g. https://api.together.xyz/v1)
3. Enter API key
4. Select a model from the provider's catalog
5. Done — Claude Code sessions now use that provider

Environment variables (for docker-compose users)

environment:
  - OPENAI_COMPAT_API_KEY=your_key
  - OPENAI_COMPAT_BASE_URL=https://api.provider.com/v1
  - OPENAI_COMPAT_MODEL=meta-llama/Llama-3.3-70B-Instruct

Providers this would unlock

Any provider exposing /v1/chat/completions, including but not limited to:

Cloud providers

Provider	Endpoint
Hyperstack AI Studio	https://console.hyperstack.cloud/ai/api/v1
Together AI	https://api.together.xyz/v1
Groq	https://api.groq.com/openai/v1
Fireworks AI	https://api.fireworks.ai/inference/v1
Mistral AI	https://api.mistral.ai/v1
DeepSeek	https://api.deepseek.com/v1

Self-hosted / local inference

Server	Example endpoint
vLLM	http://localhost:8000/v1
TGI (Text Generation Inference)	http://localhost:8080/v1
Ollama (OpenAI-compat mode)	http://localhost:11434/v1
LocalAI	http://localhost:8080/v1
LM Studio	http://localhost:1234/v1
llama.cpp server	http://localhost:8080/v1
Any custom deployment	User-defined

This means users can run open-weight models (Llama, Mistral, Qwen, DeepSeek, etc.) on their own GPU hardware — whether a local workstation, a home server, or on-prem infrastructure — and use them directly in Claude Code without any cloud dependency.

Why

• "Stop configuring. Start building." — HolyClaude's own tagline. An external proxy contradicts this.
• Cost flexibility — users without an Anthropic subscription get access to open-source models on pay-per-token pricing
• Self-hosted models — teams running models on their own hardware get first-class support with zero cloud dependency
• Fine-tuned models — teams running custom models on their own infrastructure can use them in Claude Code
• One feature, many providers — a single built-in translation layer supports the entire OpenAI-compatible ecosystem

References

• Ollama Anthropic API Compatibility — reference for the translation needed
• LiteLLM Claude Code Quickstart — existing external proxy approach
• Claude Code Third-Party Providers — upstream docs on ANTHROPIC_BASE_URL

[xxdesmus]

Also could unlock Ollama Cloud as well, right? That'd be amazing.