Your agents need to know what you did today.
Not what you remember telling them.

A timeline, not a pile of screenshots

Panopticon captures your screen in the background and an LLM turns the footage into activity cards — what you did, when, in which category, with which apps.

See what agents can read →
Deep work: sync pipeline
85 min · Xcode, GitHub · 2 distractions
Research: serverless Postgres
45 min · Arc, Neon docs
Standup prep from daily recap
25 min · Slack, Notes

Served to your agents over MCP

One endpoint, five tools: list days, read a day's timeline, search it, pull the daily recap, break down time by category. Any MCP client, one bearer token.

claude mcp add --transport http panopticon https://panopticon.decentparadox.me/mcp
> get_timeline(day: "2026-07-30")
{
  "cards": [{
    "title": "Rebuilt the sync pipeline",
    "category": "Coding / Debugging",
    "start": "9:40 AM", "end": "11:05 AM"
  }, ]
}

Synced on your schedule

Every few hours the app pushes day snapshots — cards and recaps — to your own Postgres. Replayed whole, so the cloud copy always matches the source.

How the stack fits together →
02:00 sync 3 days · 41 cards · 2 recaps
05:30 sync 3 days · 47 cards · 2 recaps
09:00 sync 3 days · 52 cards · 3 recaps
12:30 sync 3 days · 58 cards · 3 recaps
  now agent read "what did I ship today?"

Your models, your machine

Timeline generation runs on whichever LLM you already pay for — Gemini, Claude Code, Codex, or a fully local model. Recordings never leave your Mac except to the provider you chose.

Get set up in minutes →
gemini claude codex ollama lm-studio

A memory stack for personal agents

Local-first by construction

Capture, analysis and storage live in a SQLite database on your Mac. The cloud copy is a projection of it, not the source of truth.

How recording works →

Built to be read by agents

No dashboard to maintain. The data model is the product: days, cards, categories, recaps — shapes an agent can query directly.

The MCP tools →

One token, your database

A single bearer token guards ingest and reads. Data lands in your own Neon Postgres — rotate the token or drop the tables any time.

The sync protocol →

Changelog

0.3Jul 30, 2026MCP server + cloud sync to Neon
0.2Jul 30, 2026Timeline restored, review flow removed, new design system
0.1Jul 30, 2026Panopticon forked from Dayflow as a headless recorder
Full history on GitHub →

Panopticon is a personal instrument: a recorder that turns a day at the screen into structured memory — so the agents that work for you know what you're up to without being told.

Run your own →

See everything.
Tell no one twice.

Server status