Everything your agents need to do real work. Your systems, your company’s context and the agents themselves, in one place.

Garden is the layer between your company and its agents. It connects to where the work happens, writes down how the work is done, and runs agents that do it.

Garden turns how your company already works into agents that do the work.

Connections

Connect GitHub, Slack, Linear, QuickBooks, Postgres and the rest once. Garden reads the original records through each Connection, and agents reach them through Garden.

One place to connectRotating a credential does not change the Connection or anything granted on it.

Memory

Garden keeps your company’s context as plain documents: what the company does, how the work is done, who does it and where it goes wrong. Each claim cites the evidence behind it.

It says what it does not knowUnknowns are written down next to the rest, not filled in.

Discovery

Discovery agents read what you connected and keep Memory current. From it they find the workflows worth automating and say how often each one happens.

It runs continuouslyNew evidence reopens the documents and proposals that depend on it.

Agents

An agent starts as a proposal: the work, the evidence for it and the smallest access it needs. Approve it and the agent runs, on a schedule or when something happens in a Connection.

Nothing exists until you approveA proposal is data. It creates no agent, identity or grant.

Your own clients

Add Garden to Claude Code, Codex or any MCP client. The agents you already run get your company’s context and the same scoped access as agents that run in Garden.

Use one part or all of itOriginal data and Memory are each usable on their own.

Also in Garden

Skills
Reusable instructions that tell an agent how to do a kind of work. A Skill never grants access.
Sandbox and browser
Runtime capabilities an agent can be granted, the same way as a Connection.
Chat
Ask Garden about your company. The built-in assistant acts with your own access, not an agent’s.
Schedules and triggers
When an agent runs is part of its definition: on a schedule, or when something happens in a Connection.

Questions

Do I have to move my agents into Garden?

No. Add Garden to the client you already use as a remote MCP server. Agents that run in Garden and agents that run in your own client are governed the same way.

What does Discovery read?

Only the Connections granted to it. It has no access by default, like every other agent.

Does Garden create agents on its own?

No. Discovery proposes an agent, and a person with authority over the requested Connections approves it. Discovery never approves its own proposals.

What if Garden cannot see how some work is done?

It says so. Coverage and unknowns are kept with the documents, and a proposal states what it does not cover.

Get started

Start with Garden

Garden is in early access. We set it up alongside your team: connect your data, find the first workflow worth automating, and put an agent on it.

Book a call

Bring the agents you already run

Point Claude Code, Codex or any MCP client at Garden to give it your company’s context and only the access its work needs.

See MCP setup