Squeeze: your time compressed to signal
Squeeze captures the screen every minute and compresses a day of work into a brief you can read before lunch. What moved, what is stalled, what the week is telling you about the week to come. Zero input. Zero privacy cost. Runs locally, always yours.
The work you already do in the tools you already use is already recorded. Your editor, your chat, your deploy, your call. Squeeze listens to that signal, runs it through a local intelligence layer, and shows you what it means. The read is the product. The firehose stays in storage.
It feeds Ghost and the agents that run the studio. Collection, compression, understanding, then the loop closes. It learns from you when you correct it, and sharper evals compound the edge.
The problem
Everyone wants to know where their time goes. Nobody wants to track it manually. Timesheets are a tax you pay in lies. The honest answer about your week is already sitting in the systems you used, you just have no way to read it back. By the time your Friday review rolls around, the week is a blur.
What we built
Squeeze reads the work you already do and turns it into the signal you actually need. A silent menubar app that captures your screen every minute, runs each frame through a local vision model, and compresses a day into an hourly and daily brief. You get a morning view, a weekly view, and a project view. The read, not the firehose.
- Silent screen capture at one-minute intervals, running locally in your menubar
- A local vision model that buckets the frames by project, learning from your corrections
- Commitments surface from your meeting notes, so nothing you said you would do gets lost
- A nudges engine that flags stalled work, overdue items, and week-on-week drift
- A team summary and weekly digest: who moved what, what is backed up, what needs attention
- Away detection and auto-pause, so you are not penalized for lunch or a meeting
- Zero outbound network by default. Nothing leaves your machine unless you opt in
- Hide any app or window you do not want it to read
Why local and learning matter
Compress the day on the machine that lived it. The compression happens where the context lives, so what gets surfaced is what matters to you, not what matters on average. The learning loop closes every week. You correct the classifier, it tightens, the brief gets cleaner. Ghost sees cleaner signal, so the agents act on truth instead of noise. The durable edge is in the closed loop, not in the raw data.
- The raw material is already yours. We just compress it into a read
- Local intelligence means the brief is personalized to your actual work, not generic
- The feedback loop: corrections sharpen the classifier, so next week is better
- The vault loop: Squeeze feeds the vault that Ghost reads from, so the agents get better signal
- No moat in the data itself; the moat is in the loop that learns and compounds
A day of work, compressed to a brief
A full day of screen frames, captured one a minute, becomes a handful of project buckets and a two-minute read. You pour your coffee and already know what moved, what needs you, and where the day is heading. The model runs locally. The brief surfaces what matters, and the rest stays in the archive.
Your week, recorded with zero input
No timesheets, no logging, no honor system, no second app to remember. The work you already do becomes a clean record of where the week actually went, captured automatically while you work, and it never leaves your machine unless you choose to send it.
It learns from you, then feeds the agents
When you correct how a moment was bucketed, the system records it and tightens the next one. A week of small corrections becomes a week of sharper reads. Then Ghost sees the clean buckets and acts on them, so the fleet runs on signal that keeps getting better.
Squeeze is operating and running the studio daily as of May 2026. What runs below is what we ship every morning. What we are building is the deeper learning and the tighter agent integration.
- Automatic screen capture and local VLM compression
- Morning brief: what moved, what needs you, project-bucketed views
- Weekly digest and Slack summary
- Commitments tracker (pulls from meeting notes)
- Nudges: stalled, overdue, drift
- Project matcher that learns from your corrections
- Away detection and auto-pause
- Deeper project-by-day pattern learning
- Tighter Ghost integration: passing clean signal directly to the agent fleet
- Per-project vault context enrichment (combining Squeeze signal with vault history)
- Reanalyze on-demand (re-run the VLM on stored frames when the classifier improves)
a note from DaveThis is one of the bets we own. Squeeze proves the value is not in the raw firehose, it is in the compression: the read. A day of frames becomes a handful of project buckets and a brief. A week of days becomes the honest shape of where the time actually went.
It also feeds Ghost. We capture the real day automatically, the agents act on what Squeeze compressed, so the fleet runs on signal, not noise. The system learns, the edge compounds. That is the thesis made real: keep the record of the real day, grade what moves through it, and correct it when it is wrong. The corrections are the part that compounds.
If you want to own the record of your own work instead of renting it out, shoot me a note.
dave@rslvd.ai →