Connectors

One system, one connector, typed tools.

A connector is a typed, testable description of one system you want your AI to reach. Every connector follows the same lifecycle, exposes the same health and test surface, and generates its tools the same way.

Available now and coming next

Available

IMAP Email

Read folders and messages, search, flag, and move mail. Write actions are opt-in and can require approval.

Planned

REST API

Base URL, auth mode, and an explicit allow-list of routes. No blanket access to an API surface.

Planned

MySQL (read-only)

Approved tables, views, and parameterized query templates. Arbitrary SQL is never executed.

Planned

Files & Documents

Approved storage sources. Read metadata and contents where safe, search where indexing is enabled.

Every connector, the same guarantees

  • Typed configuration. Validated on save, not on first failure.
  • Secrets handled once. Encrypted at rest, never rendered back to the browser, never written to a log or an audit snapshot.
  • Read-only by default. Write capability is a deliberate, per-connector decision.
  • Standard health checks. Run on a schedule, with the last error and last-healthy time visible in the portal.
  • Consistent tool generation. Tools come from capability definitions and policy, so behaviour is predictable across connector types.

On the roadmap

The connector framework is built so new systems are additions, not rewrites. Next up:

Microsoft 365 Google Workspace SharePoint / OneDrive GitHub PostgreSQL Custom enterprise APIs

Need one that isn't listed? Tell us which system — the roadmap is driven by what customers actually ask for.