A memory Zed’s agent writes and your other machines read

Zed gives its agent good places to start from. Your personal AGENTS.md applies to every project you open, a project’s own AGENTS.md, CLAUDE.md or .rules applies to that project, and the docs call both "always-on context for the Zed Agent". The Thread History "holds all your threads across all projects", and you can restore any of them. What the agent does not have is a memory of its own: none of the sixteen built-in tools in Zed’s docs is a memory tool, so what it worked out with you on Tuesday stays in Tuesday’s thread unless somebody writes it down.

Not by itself: Zed’s agent has no memory tool among its built-in tools, so a new thread starts from the instructions you write in AGENTS.md, and an old one comes back when you restore it from the Thread History. Connect Zed to Pando and the agent can write its decisions into a bullet of your outline, which your other machines, your phone and every agent you allow can read.

Set it up

Zed keeps settings.json in ~/.config/zed on macOS and in the Zed folder under %APPDATA% on Windows, and a recent release added an entry to the Agent Panel’s options menu for adding a remote MCP server. With no Authorization header in the entry, Zed runs the MCP sign-in itself: it registers with Pando as Zed and keeps the session in the system keychain, so later startups open no browser. Pando’s tokens do not expire, so that sign-in holds until you revoke it under Agents and API keys.

Then give the agent the habit in your personal instructions, the AGENTS.md in that same folder: before assuming, look in Pando, and when something is decided, write it there with the reason.

What Pando asks when Zed signs in

Zed asks for both of Pando’s scopes, so the consent page opens with “Read and write your outline” selected, and either level starts it on your whole outline. The box beneath, ticked by default, makes a bullet under Home called “Zed's memory” for it to write in, even if you choose read only.

To keep it to one branch, open that branch, open Agents and API keys in Settings, tap Zed’s row under Your agents, and press Let it reach only “…”, the bullet you are in. From then on it reads and writes that branch and what is under it, remembers there, and reaches nothing else in your outline.

A key instead of the sign-in

Zed skips the sign-in when the entry carries an Authorization header, so an agent with a key works too, and made this way it never starts on your whole outline. Under Agents and API keys, give a new agent a handle, untick “Let it read and write this outline” and “Give it a bullet of its own to remember in”, and choose Create the agent. Its key, which begins with pf_, is shown once, on the sheet that opens, so copy it there. Until you hold the agent to a branch with the same press, it reaches nothing. Then put the key into the entry:

{
  "context_servers": {
    "pando": {
      "url": "https://pando.ink/mcp",
      "headers": { "Authorization": "Bearer <key>" }
    }
  }
}

The agent beside Zed’s own

Zed also runs external agents, and those may read their own instruction files rather than Zed’s. The docs mark the personal AGENTS.md "Not generally used" by them and say "Do not assume Zed’s instruction loader controls those agents." Give that agent Pando too, through Zed, which the docs say "can forward configured MCP servers over ACP", or through the agent’s own MCP settings, and both read and write the same outline.

What each place is for

• The project’s AGENTS.md, CLAUDE.md or .rules, for decisions about that repository. Commit it with the code, and it goes wherever the code goes.
• Your personal AGENTS.md, for how you want Zed’s agent to work in every project you open.
• The Thread History, which holds the threads of all your projects, for going back into one conversation.
• A bullet of your outline, for what the agent decides as it works that has to reach another machine, another agent, or you on your phone.

When Zed’s own places are enough

For one repository, most of the time. A decision about that code belongs in the project’s own instruction file, next to the code. Zed uses the first it finds, .rules before AGENTS.md and AGENTS.md before CLAUDE.md, and its docs note that Claude, run as an external agent, reads a CLAUDE.md natively. At one desk with one agent, the personal AGENTS.md and the Thread History cover the rest, and you should use them. The outline is for what falls between: a laptop as well as a desktop, since none of Zed’s AI docs describes a sync between machines, an external agent beside Zed’s own, or a note you want to read and correct from your phone.

What it costs

Nothing, for 1,000 bullets, with every feature included and no payment card. Zed’s docs say "Zed is free to use. You can configure AI agents with your own API keys via Use API Access." What the model costs is between you, Zed and the model’s provider, and Pando does not meter agent calls.

Deeper

• Two agents, one memory
• Keeping an agent out of the rest of your notes

Connect an agent, about two minutes · The twelve tools · Which note apps an agent can reach · Who runs this