Additional model provider templates — OpenRouter, Groq, Ollama, Anthropic

[strangeadvancedmarketing]

The Problem

engine/openclaw.template.json is currently wired for NVIDIA's API (Kimi K2.5). This is a great default — free tier, 131K context, fast — but it's the single biggest setup friction point for users who don't have an NVIDIA account or want to use a different provider.

What's Needed

Tested, working openclaw.json config blocks for additional providers. Each one should be a drop-in replacement for the model + provider block in the main template.

High priority:

• OpenRouter — broadest model selection, single API key for everything
• Groq — fastest inference, free tier, good for Llama models
• Ollama — fully local, no API key, no cost. The "air-gapped" use case.
• Anthropic — Claude as the AI running the framework (meta, but valid)

Format

Each provider config should be a standalone snippet showing:

1. The model field value
2. The provider block with baseUrl and auth pattern
3. Any provider-specific quirks (context limits, rate limits, unsupported features)
4. Which models are recommended and why

These would live in a new docs/PROVIDERS.md file and be referenced from CONFIG_REFERENCE.md.

How To Contribute

1. Get the framework running on a non-NVIDIA provider
2. Document your working config (sanitize your API key — use YOUR_API_KEY_HERE)
3. Note anything that behaved differently from the NVIDIA setup
4. Open a PR adding your provider block to docs/PROVIDERS.md

Even a single working provider config is a meaningful contribution — each one removes a setup blocker for a different group of users.

Status

• NVIDIA (Kimi K2.5) — shipped, production-validated
• OpenRouter
• Groq
• Ollama
• Anthropic