aperture-models

Auto-discovers Aperture models. Point pi at an Aperture endpoint and every model it serves you appears in /model, replacing the copy/pasted models.json block from Aperture's agent setup guide.

On load the extension fetches GET <endpoint>/api/agent-config, reads the pi provider slots Aperture renders (one per wire protocol: aperture-anthropic, aperture-responses, aperture-completions, aperture-gemini), and registers them. Opening /model re-fetches, so models added or removed on the instance show up without a restart. A brand new protocol slot takes a /reload.

Discovery failing at load (off the tailnet, wrong endpoint) registers nothing and warns once per session, on stderr in print mode; fix connectivity and /reload.

Aperture renders only model ids today, so pi's models.json defaults apply (128k context, 16k output, no reasoning flag, zero cost). Extra fields Aperture adds later pass through untouched.

Config

APERTURE_URL sets the endpoint. Without it, the extension uses the baseUrl of an aperture-* provider in models.json. Without either, the default is https://ai.corp.ts.net. A trailing /v1, as other Aperture clients use, is accepted.

Install

pi install npm:@guygrigsby/pi-aperture-models