Agentslate

Agent Slate

**Shared memory and a shared workspace for you and your coding agents.**

Agent Slate keeps context alive across sessions, machines, and agent harnesses. A new session starts with the work that came before it: recent notes, completed tasks, durable knowledge, and the memory for the current machine and project.

During the session, you and the agent work from the same state. The agent reads and writes it through MCP; you see and edit it in a web UI built for desktop and phone. Canvases, files, images, and interactive documents give you a common desk for handing work back and forth.

One small server provides the web UI, MCP endpoint, session-start hooks, and SQLite store. Run it on a machine you control and reach it over a private network.


*The nest, one of Slate's boards — canvases the agent wrote, live HTML tiles beside them. Fictional data, seeded by the built-in demo.*

Quick start

The typical setup: Slate runs as a service on an always-on machine, bound to that machine's address on a private network, and every machine you run an agent on is wired to it. The examples use [Tailscale](https://tailscale.com); any VPN or private network works — your own or your company's — so long as the address is not public. The server needs [uv](https://docs.astral.sh/uv/) and Node.js; each agent machine needs Python 3 and a clone of this repository.

Everything on one laptop works too: if you are running Claude Code or Cursor on your own machine and letting it SSH into whatever it needs, Slate can live on that same laptop — macOS or Linux. Run `./start.sh` with no arguments and it binds to `127.0.0.1:8750`, reachable only from that machine; use `http://127.0.0.1:8750` as the URL in the steps below and skip the network entirely. Keep the server running in a terminal tab, or set it up as a service — the systemd unit below on Linux, a launch agent on macOS ([Beyond