Gen3 Gt Api Published Endpoints

GT API Published Endpoints

Start Here

1. Confirm GT API is enabled for your account (gtApiEnabled).
2. Open GT API at /gt-api and select Published Endpoints.
3. Decide what external callers may put in OpenAI model: agent aliases, raw-model catalog names, or both.
4. Publish only the inference names that match your integration policy (tools, prompts, and modalities stay inside GT).
5. Create or update an API key whose allowlists include exactly those published names.
6. Validate with GET /v1/models using that key—the response must list only intended names.

Why this matters

Published endpoints are the public discovery surface for GT API. Callers never need internal agent ids or provider model strings; they use published names GT maps to configured agents or operator-curated raw-model entries. If a name is not published—or not on the key allowlist—it must not appear in /v1/models or succeed on chat routes.

In-app workspace: GT API → Published Endpoints.

Full documentation (client runbooks, compatibility matrix, upload/file routes) is in Instructions under GT API—see Client Runbooks and API Reference.

Details

Two kinds of published inference names

Kind	What callers use in model	What stays internal
Agent alias	Public alias string for a configured tenant agent (tools, base prompt, speech/vision/image settings).	Underlying provider routing and agent configuration.
Published raw-model catalog name	Operator-curated catalog entry for provider-backed inference alongside agents.	Provider credentials and routing policy.

Both kinds can appear in GET /v1/models when the presenting key’s agent allowlists and raw-model allowlists permit them. The selectors on inference keys are separate—allow one without implying the other.

Relationship to agents and models

• Build and tune agents in Agents → Agent Configuration before publishing aliases.
• Control Panel operators register provider models on Models; tenant publishing decides which catalog names external callers may discover.

Upload targets (separate from inference publishing)

Upload keys authorize dataset targets (datasets:write) or conversation-scoped uploads. Publishing inference names does not automatically grant upload access—scope upload keys and conversation linking explicitly.

Operational practices

• Publish the smallest set of names required for each integration.
• Use distinct aliases when different apps need different tool, prompt, or modality policy.
• After changing published names, re-run client GET /v1/models validation before production traffic.

Related pages

• GT API Overview
• Keys and Access
• Compatibility Matrix
• Building Agents