Escribano runs on your Mac and quietly captures what's on screen. A vision model understands it. A CLI lets your agents — or you — search, filter, and cite the evidence. Nothing leaves your machine.
A small menu-bar app watches your screen in the background. Repeats are skipped, nothing is uploaded, and you can pause it whenever you want.
Each moment is turned into a short, plain-language description of what you were doing — the tools, the files, the context. All of it stays on your machine.
Ask your agent — or the command line directly — what you were working on. Escribano returns the evidence, cleanly structured, ready to cite.
A quiet Mac app that captures and understands your work. A clean interface your agents can reach for — or you, when you need to recall.
Claude Code, Cursor, Aider — they forget the moment the session ends. Point them at Escribano's CLI and they can cite the exact moment a bug appeared, the tool you opened, the file you edited.
"What was that library I tried on Tuesday?" "When did the error first show up?" Query your own work like a database, instead of scrolling through browser history.
Stable JSON, versioned API, local SQLite. Build the agent surface you actually want — an MCP server, a shell alias, a scheduled digest — on top of a substrate you own.
A screen-capture tool should earn your trust before it earns anything else. Escribano is engineered so the question "where does my data go?" has a short answer: nowhere.
Vision model, entity extraction, search index — all on your Mac. No cloud inference, no API keys, no upload. You can unplug the network.
Emails, API tokens, passwords, and secrets are filtered out of text before it ever leaves the query layer. Default on. No flag to disable.
Raw frames never appear in query results unless you pass --images. Agents can't accidentally slurp pixels into their context window.
We don't collect your data — so we can't train on it. No crash reporter, no analytics, no phone-home. The repo is public. Verify it yourself.
Early beta users get a direct line to what we build next. Join while the recorder stabilises, tell us where it breaks, vote on what matters.
Beta users get to vote on what we build next.