Connect a custom provider to Treza

The Custom provider lets you point Treza at any OpenAI-compatible chat completions endpoint — a self-hosted model (vLLM, Ollama, LM Studio), an aggregator like OpenRouter, or another provider that mirrors the OpenAI API. Treza stores your key encrypted and uses it server-side to forward redacted requests.

What you'll need

1. An OpenAI-compatible endpoint. Treza forwards requests in the OpenAI Chat Completions format, so your upstream must accept POST /chat/completions with the same request and response shape.
2. The full upstream URL, including the path — for example https://your-host/v1/chat/completions.
3. The API key your endpoint expects. If your endpoint doesn't require one, enter any non-empty placeholder value.

Add it to Treza

1. In the Treza dashboard, open Redaction proxies → New proxy and set Upstream provider to Custom.
2. Paste your key into the Provider API key field.
3. In Upstream URL, enter your endpoint's full chat completions URL.
4. Choose which PII types to redact, then create the proxy.

If requests fail, the most common cause is an endpoint that isn't fully OpenAI-compatible — check that it accepts the standard /chat/completions request body.

Your provider key is encrypted at rest and only used to forward requests upstream. Treza never returns it to your agents or includes it in logs.