AI Providers

Path: /admin/tools/ai

Configure and verify AI providers used by Orchestrator desks and automation.

Where data comes from

Layer	Source	Notes
Ollama health	Local OLLAMA_HOST (default http://127.0.0.1:11434)	Reachability, version, and model list from the Ollama API
Ollama test	Same host via Orchestrator admin API	Sends a short prompt and returns the model reply
OpenAI / Vercel	Account or server env keys	Used when those providers are configured

Ollama setup

1. Install and run Ollama on the machine that serves models.
2. Set env on the Orchestrator server (or local .env):

bash
OLLAMA_ENABLED=true
OLLAMA_HOST=http://127.0.0.1:11434
OLLAMA_MODEL=llama3

1. Open /admin/tools/ai, confirm Connected, then use Test Ollama.

Optional cost transparency vars (OLLAMA_INPUT_COST_PER_1K, OLLAMA_OUTPUT_COST_PER_1K) only affect internal usage reporting. They do not bill a cloud meter by themselves.

Limitations

1. Ollama must be reachable from the Orchestrator process. A browser on a laptop cannot talk to a remote PC Ollama unless the host URL is network-reachable and allowed.
2. Cloud deployments need their own Ollama host or a tunnel. 127.0.0.1 on the VPS is the VPS loopback, not your desk Mac.
3. Model names must match what ollama list shows on that host.
4. Provider routing for product features depends on server config. This page is a status and smoke desk, not a full routing editor.

Related

• Desk: /admin/tools/ai
• Tools hub: /admin/tools
• Integrations: /docs/integrations