The AI passes (graph enrichment, skill generation) support OpenAI, OpenRouter and Anthropic. They do not support a local model behind an OpenAI-compatible endpoint, which is what people running Ollama, vLLM or LM Studio have — and those are exactly the people who self-host a gateway so data stays local, then discover the optional AI features send it to a vendor.
Scope
AI_BASE_URL + AI_MODEL alongside the existing provider config.- No assumption of function calling or JSON mode beyond what the prompts already need; degrade with a clear log line rather than failing.
- A line in the docs saying which local models have actually been tested.
See Discussion #480.
The AI passes (graph enrichment, skill generation) support OpenAI, OpenRouter and Anthropic. They do not support a local model behind an OpenAI-compatible endpoint, which is what people running Ollama, vLLM or LM Studio have — and those are exactly the people who self-host a gateway so data stays local, then discover the optional AI features send it to a vendor.
Scope
AI_BASE_URL+AI_MODELalongside the existing provider config.See Discussion #480.