> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.fabricgateway.ai/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.fabricgateway.ai/_mcp/server.

# What is Fabric AI Gateway

> Decide whether Fabric AI Gateway fits your team. See what it covers, how it runs, what it protects, and how to try it.

Fabric AI Gateway is Postman's gateway for AI traffic. It sits between your apps and agents and the LLMs, MCP servers, and APIs they call, so that you control access, budgets, and policy in one place and see traces, cost, and usage for your traffic. One gateway covers LLM, MCP, and API traffic, instead of a separate gateway for each. This topic is for the people who decide whether to adopt it: architects, security reviewers, and engineering leads. It answers the questions that you need answered before a proof of concept.

## What it covers

Callers reach the gateway over HTTP, including the OpenAI and Anthropic APIs, or over MCP. The gateway routes each request to an upstream: an LLM from any of 70+ providers, an MCP server, or a REST API.

The data plane provides the following capabilities:

* **Auto-failover** — Circuit breaker and retry logic. To control failover on a route, see [Fail over across providers with the LLM Router](/configure/routing/failover).
* **Rate limits and budgets** — Cap requests or LLM tokens per caller in a time window, and cap LLM spend for an organization or a namespace.
* **Observability** — Cost, latency, and audit log tracking.
* **Secret isolation** — The gateway adds the upstream credentials, so callers don't hold provider keys. The exception is a route where you let callers send their own keys, such as one that uses the [Passthrough plugin](/reference/plugins/passthrough).
* **Guardrails** — PII redaction on LLM prompts and MCP tool arguments, access control per service and tool, and an audit trail.

On the Enterprise plan, you run the data plane in your own Kubernetes cluster. On the Free plan, Postman runs it for you. For details, see [Choose a deployment model](/evaluate/deployment-models).

## Find the answer you need

| Your question                                                  | Where to look                                                             |
| -------------------------------------------------------------- | ------------------------------------------------------------------------- |
| How does it work?                                              | [How Fabric works](/evaluate/architecture)                                |
| Can it run in my environment, and what does each plan include? | [Choose a deployment model](/evaluate/deployment-models)                  |
| Where does my data go, and who can see my secrets?             | [Review security and data handling](/evaluate/security-and-data-handling) |
| What are the current limitations?                              | [Know the current limitations](/evaluate/limits)                          |
| How do I try it?                                               | [Try Fabric](/evaluate/try-fabric)                                        |

## See who does what

After you adopt Fabric, three roles use it: consumers who call the gateway, operators who configure it, and infrastructure engineers who run the data plane. For details, see [Understand who uses Fabric](/evaluate/who-uses-fabric).

## Next steps

* [Try Fabric](/evaluate/try-fabric)
* [How Fabric works](/evaluate/architecture)