Gen3 Gt Api Overview - GT-Edge-AI/GT-AI-OS GitHub Wiki

Start Here

Use the tenant GT API base URL (https://<tenant-host>/api/tenant) with a scoped bearer key. Validate GET /v1/models before enabling chat, embeddings, uploads, or GT conversation headers.

Why this matters

External tools only see published inference and upload aliases. GT API keeps internal agent and dataset identifiers private while exposing OpenAI-shaped routes for compatible clients.

Details

Open the GT API workspace (/gt-api) for API Keys and Published Endpoints. Full compatibility matrix, per-client runbooks, and API reference live in this Instructions section.

Quickstart bundle

Base URL: https://<tenant-host>/api/tenant
Authorization: Bearer gtak_...

List published models:
curl -sS \
  -H "Authorization: Bearer <YOUR_GT_API_KEY>" \
  "https://<tenant-host>/api/tenant/v1/models"

Chat completion:
curl -sS \
  -H "Authorization: Bearer <YOUR_GT_API_KEY>" \
  -H "Content-Type: application/json" \
  "https://<tenant-host>/api/tenant/v1/chat/completions" \
  -d '{
    "model": "agent-alias",
    "messages": [
      { "role": "user", "content": "Summarize the attached mission brief." }
    ]
  }'

Getting started notices

Simple inference-only rollout

Use this path when an integration only needs OpenAI-compatible model discovery and chat completion.

  • Publish inference targets in Published Endpoints: agent aliases for configured agents, and published raw-model catalog entries when callers should use those names in model instead of an agent alias.
  • Create an inference key limited to the exact agent aliases and raw-model targets (and inference scopes) that caller should reach; add prompt guardrail modules or extra prompt text on the key when policy requires it.
  • Hand the caller only the bearer key and the /api/tenant/v1/* base URL.

Advanced conversation plus uploads

Use this path when the client app needs one conversation-facing credential that can both chat and attach files.

  • Create reusable inference and upload keys first so their allowlists stay independently auditable.
  • Create the conversation-facing key from API Keys, linking it to those reusable runtime keys.
  • Publish only the inference names the caller should see after authentication—agent aliases, raw-model catalog names, or both—then create a GT conversation explicitly before sending follow-up chat turns and file flows with X-GT-Conversation-Id.

Custom OpenAI-compatible inference URL

Use this path when a third-party product is not listed in the matrix but lets you configure a custom OpenAI-compatible base URL and bearer key.

  • Point the client at the tenant GT API base ending in /api/tenant, then validate GET /v1/models before enabling higher-level features.
  • Start with a narrow inference key and at least one published inference name (agent alias or raw-model catalog entry), then add embeddings, uploads, or GT conversations only if the product truly needs them.
  • If the product only supports vendor-native APIs and cannot target an arbitrary OpenAI-compatible URL, GT API is not a direct fit without a proxy layer.