Artemii
Sign in Open Artemii

Developers

Build with Artemii.

An OpenAI-compatible API, a drop-in assistant panel and an actions protocol, so any Mayfair app can add Artemii.

What you can build

Call the API

Send chat requests from your server with a service key, or from a signed-in browser with a Mayfair ID token. Most OpenAI client libraries work by changing the base URL.

Embed the assistant

One script tag adds a copilot panel to your app. It knows what the person is looking at and can run the actions you give it, after they approve.

Give Artemii app tools

Publish a manifest and one endpoint, and people can ask Artemii to search their mail or create an event, as themselves.

Share memory

Read and write the same per-person memory that Artemii uses, with the person's switches and a filter for sensitive data.

The basics

  • Base URL. <span data-api-url>https://mayfair-ai.crelloo.workers.dev</span>. Every endpoint lives under /v1.
  • Models. The public ids arlo, lucian, orion and curzon. See Models for what each is for.
  • Identity. The Artemii identity and voice are always applied. Your own system messages add to them; they cannot remove or override them.
  • Stateless. A call saves nothing: no chats, messages, files, memory or artifacts. Request text is not logged by the gateway; usage counts are.
  • Text only. Messages are text. Images are not supported in requests.

Getting access

Service keys are created by a Mayfair administrator in the admin console, and apps that sign people in with Mayfair ID are added to the gateway's allowed audiences and origins. If you are building a Mayfair app and need either, email hello@mayfairmail.com with the app name and what it will do. There is no self-serve sign-up for keys yet.

Versioning

The API is versioned by path (/v1). Inside /v1 we only add: new optional request fields, new response fields, new models and new error codes, so clients must ignore unknown fields. Anything breaking becomes /v2, and /v1 keeps working alongside it with advance notice before it is retired. Model ids are stable. A model's version, shown by GET /v1/models, changes when it is tuned.

Where to go next