Does my screen data get sent to Zed or its model providers?

The memory itself stays on your machine. What an agent actually pulls becomes part of that conversation. If the agent runs in someone else’s cloud, that conversation goes there. Connect only the agents you trust, and ask them for what you are willing to put in that thread.

Context servers or skills?

Skills. They tell the agent when to search, which is the part a bare tool list cannot do. Add LUCI under context_servers only if you want the tools without the skill files.

Do I need to keep LUCI running?

For live history, yes. If LUCI is quit, the luci skill reads your Life folder instead. The memory only grows while LUCI is capturing.

Coding agent

Give Zed's Agent Memory of Your Whole Day

The agent panel in Zed is quick, and it starts every thread empty. Install the official LUCI skills and it can search what was on this screen and what was said in your calls, from a memory that is built and stored here.

See all integrations

Zed's agent reads the project and the files you point it at. Everything else about your day is invisible to it: the browser tabs, the terminal windows, the calls. So you spend the first minute of every thread typing context you already had.

The official LUCI skills fix that. Zed reads skills from ~/.agents/skills globally and .agents/skills inside a worktree, one folder per skill, each holding a SKILL.md. Put the two LUCI skills there and the agent knows when to look something up.

~/.agents/skills is shared ground. pi and Amp read it too, so a single copy in that folder covers three agents, and the same local memory answers all of them.

What Zed can do with your memory

One folder, three agents

~/.agents/skills is read by Zed, pi, and Amp. Install once and all three can search your memory.

Exact strings and vague memories

Search the error text you remember word for word, or the article you only remember the topic of.

What was said, on the record

Microphone and system audio are transcribed on this machine, so the agent can quote the call instead of guessing at it.

MCP if you want it

Zed calls MCP servers context servers. LUCI connects either way, and skills stay the path we recommend.

Try asking

Questions that work on day one

What was I reading before I opened this file?
Search my screen memory for the migration error from this morning
Find where we agreed on the API shape in the call
What was I doing between 9am and 11am?

Set up Zed

01
Turn on Screen Memory in LUCI
Install LUCI on your Mac or Windows machine, enable Screen Memory, and leave the app running.
02
Install the official skills
Put the two skill folders in ~/.agents/skills, or in .agents/skills inside a project. Each skill is a direct child of that folder, because Zed does not look inside nested subfolders: https://github.com/Memories-ai-labs/Luci-skills
```
npx skills add Memories-ai-labs/Luci-skills -g
```
03
Optional: connect over MCP
Run zed: open settings file and add LUCI under context_servers, using the URL and token from the Screen Memory panel.
```
{
  "context_servers": {
    "luci": {
      "url": "<url>",
      "headers": { "Authorization": "Bearer <token>" }
    }
  }
}
```

Where Zed looks for it

The two skill folders are ordinary files you own. Drop them where this tool reads and it finds them on the next session.

Global skills folder: ~/.agents/skills/
Project skills folder: <worktree>/.agents/skills/
Optional MCP key: context_servers in settings.json
Settings file: ~/.config/zed/settings.json, %APPDATA%\Zed\settings.json

Frequently asked questions

Check three things: each skill is a direct child of the skills folder rather than nested in a subfolder, the file is named SKILL.md in capitals, and the frontmatter carries a name and a description. If no skills load at all, update Zed. Agent skills replaced the older rules library in a recent release.