The Gateway API is the public surface of ai_gateway projects. Two routes:

Route	Use
POST /api/{uuid}/{slug}	Direct call to a single endpoint by slug.
POST /api/{uuid}/chat/completions	Generic dispatcher — request body picks the endpoint by slug.

Both require chat scope.

The two routes are functionally equivalent; the second is more familiar to OpenAI users (one URL, multiple endpoints) and is what you’d use if a downstream tool expects a single OpenAI-style URL but you want endpoint routing.

Route 1: Direct slug

Request

POST /api/{uuid}/{slug}
Content-Type: application/json
Authorization: Bearer pg_live_...

Body — single message:

{ "message": "Explain quantum computing." }

Or full conversation:

{
  "messages": [
    { "role": "system", "content": "You are a teacher." },
    { "role": "user", "content": "What's a qubit?" }
  ]
}

Optional fields:

Field	Type	Default	Notes
stream	bool	false	Honoured if endpoint has streaming enabled.
session_id	string (uuid)	absent	Resume an existing session.

Response (non-streaming)

{
  "ok": true,
  "id": "chatcmpl-abcd1234",
  "model": "gpt-4o-mini",
  "content": "A qubit is the quantum analogue of a classical bit...",
  "finish_reason": "stop",
  "usage": {
    "prompt_tokens": 38,
    "completion_tokens": 87,
    "total_tokens": 125
  },
  "meta": {
    "request_id": "0e2f...c4",
    "session_id": "..."   // only when session is created/resumed
  }
}

Response (streaming)

OpenAI-style SSE chunks. See Streaming.

curl

curl -X POST $PG_URL/api/$PG_UUID/$PG_SLUG \
  -H "Authorization: Bearer $PG_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"message": "Explain quantum computing."}'

Python

import os, requests

resp = requests.post(
    f"{os.environ['PG_URL']}/api/{os.environ['PG_UUID']}/{os.environ['PG_SLUG']}",
    headers={"Authorization": f"Bearer {os.environ['PG_TOKEN']}"},
    json={"message": "Explain quantum computing."},
    timeout=120,
)
resp.raise_for_status()
print(resp.json()["content"])

Node.js

const r = await fetch(`${process.env.PG_URL}/api/${process.env.PG_UUID}/${process.env.PG_SLUG}`, {
    method: 'POST',
    headers: {
        'Authorization': `Bearer ${process.env.PG_TOKEN}`,
        'Content-Type': 'application/json',
    },
    body: JSON.stringify({ message: 'Explain quantum computing.' }),
});
console.log((await r.json()).content);

Route 2: /chat/completions dispatcher

Same as Route 1, but the endpoint slug goes in the body:

{
  "endpoint": "summarize",
  "messages": [
    { "role": "user", "content": "Long text..." }
  ]
}

Useful when you have a tool that talks to a single URL (POST .../chat/completions) but you want to route across multiple endpoints in the project.

curl

curl -X POST $PG_URL/api/$PG_UUID/chat/completions \
  -H "Authorization: Bearer $PG_TOKEN" \
  -d '{
    "endpoint": "summarize",
    "messages": [{"role":"user","content":"Long text..."}]
  }'

The response shape is identical to Route 1.

Behaviour reference

Situation	Status	Body
Success	200	{ok: true, ...}
No bearer token	401	{ok: false, error: "Missing Bearer token."}
Token from different project	403	{ok: false, error: "Token does not belong to this project."}
Wrong scope	403	scope mismatch message
Endpoint slug not found / inactive	404	endpoint-not-found message
Per-minute rate limit hit	429	+ Retry-After header
Per-request token cap exceeded	422	budget message
Monthly budget exhausted	422	budget message
Guardrail blocked	422	which guardrail
Input schema invalid	422	validator detail
Provider error after failover	502	provider error
Output schema invalid	502	validator detail
Wrong project type for this surface	400	type mismatch

Pipeline order

Request → auth → rate limit → budget → guardrails → input schema → prompt apply → provider call (with failover) → output schema → log → response.

So a request that fails an early step never pays the cost of later steps.

Common patterns

Single-shot chat

Just pass message. PromptGate wraps it as [{role: "user", content: ...}].

Multi-turn (no session)

Pass the full messages history. The client manages the transcript.

Multi-turn (with session)

Pass session_id + the new user message. The gateway stores the history and prepends it next time.

Streaming

Pass "stream": true. Endpoint has to have streaming_enabled=true. Response is SSE.

---

Next: Wrapper API.

---

© Akyros Labs LLC. All rights reserved.