UbiquityOS / AI Gateway
Developer portal

AI.UBQ.FI Developer Portal

Integrate approved UbiquityOS applications and agents with the OpenAI-compatible UbiquityOS AI Gateway.

OpenAPI contract Agent instructions

Quickstart

Set an approved UbiquityOS gateway token in UOS_AI_TOKEN. Configure an OpenAI-compatible client with https://ai.ubq.fi/v1 as its base URL. The gateway accepts bearer authentication and does not forward a client bearer token upstream.

export UOS_AI_TOKEN="..."

curl -sS https://ai.ubq.fi/v1/models \
  -H "Authorization: Bearer $UOS_AI_TOKEN"

Always query GET /v1/models before selecting a model. The returned catalog is the source of truth for the account and runtime; do not assume that a public OpenAI model alias is available here.

API keys and authentication

This portal documents an authenticated service, not a self-service key issuer. API keys are provisioned through the authorized UbiquityOS operator workflow. If an agent or application does not already have an approved token, it must stop and ask its operator for access rather than requesting a secret in chat, logging it, or adding it to source code. Administrators use the separate protected administration flow to create and revoke keys.

Endpoints and function tools

Use POST /v1/chat/completions for the Chat Completions API or POST /v1/responses for the Responses API. Both support streaming and documented function tools. The OpenAPI 3.1 contract contains typed request fields, unique operation identifiers, response envelopes, authentication, and rate-limit headers for tool-aware clients.

Parse JSON error responses as an error object with a message, type, and code. A matching If-None-Match request to GET /v1/models?client_version=X.Y.Z can return 304 Not Modified with no body; reuse the cached catalog instead of parsing a response body. For a 429 error, honor Retry-After when present. When an enforced gateway key window is authoritative, the response also includes RateLimit fields for self-throttling.

Interactive sandbox

The Chat playground is an authenticated browser sandbox for testing a token that has already been issued to you. It is useful for a small manual request before automating an integration. It does not create credentials, expose server-side provider credentials, or replace the API contract.

Agent discovery

Autonomous agents should read /llms.txt, then the complete guide and OpenAPI document. Use this service for approved OpenAI-compatible inference, dynamic model discovery, and function calls. Do not use it for key administration, private health probes, or unsupported product actions.