One capture surface for your Mac
01 Dictation
Talking is 4x faster than typing
Hold the Globe key and say the thing. Release, and the words land in whatever field the cursor is in — notes, email, chat, a terminal.
- Speech-to-text runs on your Mac — whisper.cpp or Apple's Speech framework.
- No supported-app list. If the cursor is blinking there, the words land there.
- A cleanup pass tidies the words before they land.
- Nothing joins your call — no bot in the participant list, no link to hand out.
- The audio is transcribed on your Mac and the transcript lands in your account as plain `.md` and `.json`. `grep` works on it, because it is a file.
- Who said what, what was decided, where to pick it up.
- The screen is read on your Mac. What is kept is the text, a timestamp, and the name of the app in front.
- The picture itself is discarded after the read.
- No model is called on that cadence, so you can leave it on all day.
- Every answer names where it came from: the meeting, the timestamp, the file.
- Both citations are ordinary files in your account. Open them, export them, delete them.
- The same question works from your own agent. That is the next panel.
- Five tools, including `run` — a shell jailed to your own data slice, with no other users, no secrets and no network.
- Auth is a bearer token you can revoke. Claude Code, Codex and Cursor all work today.
- When your own agent reads the surface this way, the model is yours — those tokens are on your plan, not ours.
- Secrets are redacted on your Mac API keys, tokens, and card numbers become `[REDACTED:type:length]` before any capture is stored or uploaded — nothing recoverable, no way to get them back.
- Password managers are not captured When one is frontmost the frame is skipped entirely, so there is nothing to redact. In the current release.
- Raw audio is kept on your Mac (Surface 0.32.0+) The recording is kept on your Mac until transcription succeeds; the transcript is what syncs. Audio recorded by earlier versions was uploaded and remains in your account until you delete it.
- Encrypted at rest, per user Your captures are sealed with AES-256-GCM under your own data key, wrapped by a master key held in a separate vault.
- A named model provider Chat runs on Anthropic's Claude, and under Anthropic's API terms your content is not used to train models by default.
- Pause is one click, and recording reads red
dictation — hold to talk
hold the Globe key to talk
Notes — planning
landed
02 Meetings
Who said what, written down as it happens
Nothing joins your call and there is nothing to start. Recording is running and the transcript is already there when you hang up.
meeting — in progress
rec 00:14:22 transcribing…
03 Screen history
“Never forget how you configured that server.”
That is a real question Screen History answers. What was on your screen is read on your Mac and kept as timestamped text — so the terminal from three weeks ago is a file you can search, not a memory you have to reconstruct.
screen — one read, one display
OCR on your Mac
text — kept ssh -L 8100:localhost:8100 admin@prod-box 3 weeks ago · Terminal
image — discarded
04 Chat
Ask your Mac what happened
Chat reads the capture files in your account and answers with the receipts — the meeting transcript and the screen capture it drew from.
digitalsurfacelabs.com/u/you live
You
Surface
cited from
audio/2026/07/auto\_1030\_planning-call.md meeting ocr/2026/07/28/1042\_Display\_1.json screen
Ordinary files in your account. Open them, export them, delete them.
Ask anything
05 Your agent
Your own agent reads the same files
Claude Code and Codex read your capture files straight off your disk — transcripts and screen text are plain `.md` and `.json`. Cursor connects to the hosted MCP server with one line and a bearer header.
zsh — \~ dsl mcp
$ claude mcp add --transport http dsl https://digitalsurfacelabs.com/mcp --header "Authorization: Bearer $TOK"
Added HTTP MCP server dsl · 5 tools
\>
\[1\] meeting · planning call, Tue 10:30 · audio/2026/07/auto\_1030\_planning-call.md
\[2\] screen · Tue 10:42 · ocr/2026/07/28/1042\_Display\_1.json
— Built for how you work
Pick the page that sounds like your week
/for/engineers Engineers Never lose your work. Meetings, dictation and screen history as files your own agent can grep. See the page /for/designers Product designers Every review, every decision, every screen you had open — recalled the moment you ask. See the page /for/scrum-masters Scrum masters Who said what, what was decided, who owns it — written down as it happens. See the page
— Security
How we keep it secure
Six claims, each one checkable.
Read the full privacy & security page— every claim there traces to shipped code.
— Pricing
Setup with Onboarding Expert
Your work, your workflows, and all your questions answered. Paid plans include a monthly pool of model usage, one budget across Chat, Meetings, and Dictation.
Book a 25-minute onboarding and leave with it running on your real work. We install the capture app together, get dictation landing at your cursor, and point your own agent at your surface before you hang up.
Full plan details on the pricing page.