Computer historyfor AI agents

Your coding agent can read every file in the repo and still has no idea what you did yesterday. LUCI keeps a local history of your machine — screen, meetings, and the work that happens around the code — and lets Claude Code, Cursor, and Codex query it. Free on Mac and Windows.

What is computer history for AI agents?

Computer history for AI agents is a record of what actually happened on your machine, kept in a form an agent can search: what you had open, what you read, what you changed, and what was said on your calls. LUCI is a free desktop app for Mac and Windows from Memories AI that remembers your screen and meetings on your own device and gives that memory to AI agents like Claude Code, Cursor, and Codex. The repository tells an agent what the code is; computer history tells it what the work was. Without it, the person holding all the context is still you, and you spend the session typing it back in.

Four things your agent cannot see

The work happens outside the repo

The reason for the workaround is in a thread. The number the API has to match came off a dashboard. The reason you reverted is in a call recording. Git sees none of it, so the agent asks and you become the courier. LUCI keeps your whole day on the computer — meetings, email, code, everything in between — in one place the agent can query.

Sessions end; the project does not

Every new session starts from zero and you re-explain the same three constraints. A local history of your own machine survives the context window, so the agent picks up where the work left off rather than where the transcript did.

Decisions are spoken, not written

Every word — yours and theirs — in meetings and calls, transcribed on-device and linked to what was on screen while it was said. When the agent needs to know why the schema changed, the answer is the sentence somebody said on Thursday, not a guess from the diff.

Records beat summaries

A distilled memory keeps what a summarizer thought was durable and drops the rest, which fails exactly when you need the literal error string or the exact version number. LUCI keeps the record underneath the summary, so the agent can search broadly and then drill into the specific moment.

What it looks like in practice

Works with Claude, Cursor, and Codex — LUCI connects itself, nothing to set up. From then on the history is just another thing your agent can consult, in the same session where you are already working.

  • “What broke the staging deploy on Wednesday?” The agent finds the terminal output you were staring at, plus the call where somebody said the migration had already run.
  • “Write the PR description for this branch.” It pulls the review comments you read, the ticket you had open, and the constraint the designer raised on the call, instead of paraphrasing the diff.
  • “What was that error string?” Not a summary of the error. The string, from the frame it appeared in.
  • “Catch me up on this project.” Two weeks of work across editor, browser, and calls, in the order it happened.

At the end of each day, your agent distills what happened into your Life folder: plain files on your disk that you can read, edit, or throw away. Setup details are in the docs.

Tools that give agents a history of your machine

Five options, compared on what they capture and who can read it. Facts come from each vendor's public documentation; a dash means we could not source the answer.

ToolPlatformScreen captureMeeting audioWhere data is processedWorks with AI agentsPrice
LUCIMac & WindowsContinuous, with on-device understandingYours and theirs, in meetings and callsOn your device, encrypted at restClaude, Cursor, CodexFree
PiecesMac, Windows, Linux, IDEsSamples workflow context from apps you useNoOn-deviceClaude, Cursor, CodexFree tier; Pro $18.99/mo
ScreenpipeMac, Windows, Linux24/7 capture with OCRYesOn-deviceNoFree tier; $21/mo or $250/yr
OpenAI Codex ChroniclemacOS onlyScreen, scoped to coding workScreenshots sent to OpenAICodex onlyRequires ChatGPT Pro
Microsoft RecallWindows 11 24H2+, Copilot+ PC onlySnapshots every few secondsNo — does not record audioOn-device, Copilot+ PC requiredNoFree with a Copilot+ PC

Frequently asked questions

What is computer history for AI agents?

It is a record of what actually happened on your machine — what you had open, what you read, what you changed, what was said on calls — kept in a form an AI agent can query. The repo tells an agent what the code is. Computer history tells it what the last two weeks of work were about.

How do I give Claude Code memory of my computer?

Install LUCI and let it capture. Works with Claude, Cursor, and Codex — LUCI connects itself, nothing to set up. After that, 'what did we decide about the retry logic on Tuesday' is a prompt in the agent you already run, not a hunt through Slack.

Is this different from CLAUDE.md or project rules?

Yes, and they work well together. A rules file is something you write and maintain by hand: conventions, preferences, standing instructions. Computer history is the part nobody wants to maintain — the running record of the work itself, which stays current without anyone editing it.

Does my whole history get sent to the model?

No. The agent asks for what it needs and gets back the matching slices, not the archive. What it retrieves becomes part of that conversation, which is why you choose which agents you connect and what you ask them to fetch. The memory itself stays on this machine.

What about secrets in my terminal and browser?

Cards, keys, and passwords are blacked out before anything is saved, and you can leave specific apps and windows out of scope entirely: a password manager, a staging console, a personal inbox.

Does it work on a machine with no NPU?

Yes. LUCI runs on any modern Mac or Windows machine and uses your NPU or GPU when there is one. Microsoft Recall, by contrast, requires a Copilot+ PC with a 40-TOPS NPU and 16 GB of RAM.

Stop pasting context into your agent

A local history of your machine that Claude Code, Cursor, and Codex can read.
Free on Mac and Windows. Encrypted at rest.