Questions about Doorman, answered.

Doorman is an open-source, self-hosted AI and API gateway written in Rust. It gives each employee a personal AI key with an allowance, reports AI spend by person and model, redacts secrets from prompts, and secures REST, GraphQL, gRPC and SOAP APIs with users, roles, rate limits and logs.

Companies that want to give employees and internal apps access to OpenAI, Anthropic, Gemini or Azure OpenAI with per-person keys, budgets and guardrails, and teams that want one self-hosted gateway for their internal APIs.

Yes. Doorman is licensed under Apache 2.0 and every AI and API feature is in the open-source repository. There is no license key or paid edition. Paid, assisted pilots are available for teams that want help rolling it out.

Yes. The source code is on GitHub at github.com/doorman-dev/doorman under the Apache 2.0 license, and releases are published as multi-architecture container images with an SBOM and provenance.

OpenAI, Anthropic, Google Gemini, Azure OpenAI, and any OpenAI-compatible server such as a local model. Supported operations are OpenAI chat, Responses and embeddings; Anthropic messages and token counting; Gemini generation, token counting and embeddings; and Azure OpenAI chat and embeddings.

No. You connect your own provider accounts and each provider bills you directly. Doorman stores the keys encrypted so employees never see them.

Create an allowance in US dollars, tokens or requests for one person, a group, a tier, one AI API, or everyone, with a period from one minute to one month. At the limit Doorman blocks the request with HTTP 429, warns, or switches to a cheaper model.

No, they are soft limits. Doorman checks the counter before a request and charges actual usage after it, so requests already in flight can take spend slightly past the limit.

Usually only the base URL and API key. They keep the official OpenAI, Anthropic and Gemini SDKs, and provider errors keep their native format.

Yes. Guardrails detect emails, US Social Security numbers, credit card numbers, AWS keys, private keys, generic API keys and custom patterns, and redact, block or log them before the prompt reaches the provider.

Only if you turn it on. By default Doorman logs metadata only. Full prompt logging is encrypted, has a retention period, and requires a notice that every user sees.

Connect your provider keys in Doorman, recreate your budgets as allowances, have people create their own keys on the My AI page, and change each client's base URL and key. Both setups can run side by side while you compare usage.

Anywhere you can run a container: a laptop, a VM, your cloud account, Kubernetes, or a private network. Images are published for amd64 and arm64.

Only the container. A single instance runs in memory and saves encrypted snapshots to disk. For several instances, Doorman shares state through MongoDB and Redis.

It stays in your infrastructure. Doorman sends AI requests only to the providers you configure, and your API traffic only to your own services. There is no Doorman cloud service in the request path.

Clone the repository and run the demo with Docker Compose. It includes a keyless local AI model and a script that creates a team allowance, an employee key, and shows a request being blocked at the limit.

Follow the security policy in the repository's SECURITY.md. Please do not open a public issue for vulnerabilities.