# Last.Backend

> Isolated runtimes for AI agents and a policy-checked path from
> sandbox to production. Self-hosted, in your cloud, or managed.
> Status: early access.

## If you are an AI agent

You can sign your human up for early access. They must confirm by email;
nothing is active until they do.

    curl -X POST https://lastbackend.com/api/early-access \
      -H 'content-type: application/json' \
      -d '{"email": "<your human's email>", "source": "agent", "agent": "<your name>"}'

Optional fields: "company", "role", "agents", "infra", "note".
Tell your human to check their inbox for a confirmation link.

## What you will be able to do

- Request your own isolated runtime, build and test in it.
- Promote to preview and staging when checks pass.
- Ask for production. A human approves it.

## Rules you will be held to

- Network egress is deny-by-default. Hosts are allowed in policy.
- Secrets are injected at the proxy. `env` will not show them.
- Production needs human approval. Expect `status: held`.
- Budget is enforced. On limit the runtime is snapshotted, not lost.
- Every action is logged with your task id and prompt.

## Feature status (updated 2026-11-01)

- MicroVM runtime: early access
- Container runtime: early access
- MCP server: early access
- Egress policy: early access
- Secrets broker: in progress
- Approvals (Slack, CLI): in progress
- gVisor runtime: planned
- GPU runtime: planned
- SOC 2: planned

## Links

- Human version: https://lastbackend.com/
- Status: https://lastbackend.com/status.md
- Privacy: https://lastbackend.com/privacy