Documentation draft. Align model IDs with the current admin UI.
The bbxGrid AI agent and AI jobs call LLMs via an OpenAI-compatible API or a separate Anthropic (Claude) key. Keys: global (server) · per-user · (by policy) per-organization. Billing: plan quotas + PAYG.
What you can connect
| Provider | How | Example models |
|---|---|---|
| OpenAI | API key in AI settings; model id as in the OpenAI console | gpt-4.1-mini (code default), gpt-4.1, other chat models on the account |
| Azure OpenAI | OpenAI-compatible deployment base_url + Azure key | Your deployment name as model |
| Groq / Together / Fireworks / other OpenAI-compatible | Provider key + their API base_url | Models from their catalog (check JSON mode support) |
| Ollama / local gateway | base_url to your host (e.g. http://host:11434/v1) + dummy/local key if required | Local models behind an OpenAI-compatible shim |
| Anthropic Claude | Separate Anthropic API key (global or per-user); provider resolved from model | Claude family (current ids in Anthropic console) |
How to connect (generic recipe)
- bbxGrid admin → AI settings (or user profile for a personal key).
- Pick provider / paste API key.
- Set base_url when needed (Azure, Groq, Ollama…).
- Set model for the agent and/or AI jobs.
- Smoke-test the agent; watch plan quotas on Pricing.
Key scopes
- Global — server default.
- User — operator BYOK.
- Org — when enabled by policy (client cost isolation).
Limits (draft)
- Needs compatible chat/completions (or Anthropic messages) and stable JSON from the agent.
- Vision / camera providers are a separate module — not the LLM agent.
- Local Free: your own keys only, no cloud AI billing.