Quickstart

aigw is an OpenAI-compatible, governed AI gateway: point any OpenAI SDK at one base URL, use a virtual key, and reach every model behind a single contract - with per-key budgets, DLP, and a full audit log.

Zero to your first completion in three steps. Everything past step 1 is copy-paste, then governance is a 60-second add-on.

1. Get a key

Virtual keys are created in the console (you need a browser once for the magic-link sign-in):

console.aigw.app → sign in.
API keys → New key.
Set a monthly budget (in cents) and, optionally, an allow-list of models.
Copy the aigw_… secret - it’s shown exactly once.

A virtual key is not a provider key. It’s your key, scoped and capped by you; aigw resolves it to the right upstream provider on each request.

2. Point your OpenAI client here

The base URL is https://gateway.aigw.app/v1 (or your own gateway host). Use the virtual key as the bearer. Any OpenAI-compatible SDK works unchanged - only base_url and api_key change.

curl:

curl https://gateway.aigw.app/v1/chat/completions \
  -H "Authorization: Bearer aigw_…" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt",
    "messages": [{"role": "user", "content": "Hello from aigw"}]
  }'

OpenAI Python SDK:

from openai import OpenAI

client = OpenAI(
    base_url="https://gateway.aigw.app/v1",
    api_key="aigw_…",
)

resp = client.chat.completions.create(
    model="claude-sonnet-4-6",
    messages=[{"role": "user", "content": "Hello from aigw"}],
)
print(resp.choices[0].message.content)

OpenAI Node SDK:

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://gateway.aigw.app/v1",
  apiKey: "aigw_…",
});

const resp = await client.chat.completions.create({
  model: "gpt",
  messages: [{ role: "user", content: "Hello from aigw" }],
});
console.log(resp.choices[0].message.content);

That is the whole integration. Streaming, embeddings (POST /v1/embeddings), and the model list (GET /v1/models) all work the same way.

3. Switch models without changing code

Change the model string to route to a different provider. The same request shape reaches OpenAI-, Anthropic-, and OpenAI-compatible providers - aigw translates as needed.

resp = client.chat.completions.create(
    model="mistral-large",        # was claude-sonnet-4-6
    messages=[{"role": "user", "content": "Same code, different model."}],
)

Add governance in 60 seconds

Everything above already runs through per-key budgets, DLP redaction, and a sealed audit log. To govern what your agents may do, add these in the console - each takes about a minute:

Register a tool. In Tools, add an MCP server by URL. aigw introspects the tools it exposes so you govern real tool and argument names, not blind globs.
Register an agent. In Agents, add an agent-to-agent (A2A) endpoint. aigw reads its capability card so delegation is governed by what the agent can actually do.
Set a guardrail or approval. On any tool or agent, set a rule to allow, deny, or approve - argument-level constraints and rate limits included. approve holds the call for a human decision.

See the full walkthrough in agentic governance - tools & A2A.

What happens on every request