About Bezalel

Bezalel is a capability plane for AI agents: one deployable service that holds your connected apps and a shared memory, plus email, money, texting, a computer, and code sandboxes, and serves them to any agent runtime over the Model Context Protocol.

Why it exists

Every new agent framework means re-wiring the same capabilities inside a different extension model. The tools are the durable asset; the framework is the replaceable head. Bezalel fixes that inversion: capabilities live in one service with its own state, credentials, and event routing, and any agent connects with exactly two values — a URL and a bearer token. Switching from one framework to the next stops being a migration and becomes a config change.

What makes it different

Bezalel keeps the apps you connect and one shared memory behind a single MCP endpoint, so every agent you allow works from the same connections and the same memory. Each agent gets its own token and only the capabilities you turn on for it. The plane checks every call, and the Activity page shows what each agent did with your apps, email, memory, texts, computer, and sandboxes.

Most tool servers only answer requests. Bezalel handles the inbound side too: a text from your paired iMessage number or a mail to your agent’s inbox becomes an event that can wake an agent. Beyond your apps, it holds what an agent needs to act for you: an inbox, a texting line, card purchases you approve, a cloud computer, and sandboxes.

How it is run

The hosted service is one multi-tenant plane. Each account is a separate tenant: its data carries its owner id, and its agents share state with each other, never with another account. The plane is also self-hostable: it ships as one container image you can run wherever you like, with your own database and provider credentials.

The name

Bezalel is the Bible’s craftsman — called by name and filled “with skill, ability and knowledge” to build the instruments of the Tabernacle. A fitting patron for the toolkit an agent picks up to do real work.

Questions? See contact — or, if you are an agent, start at llms.txt.