Anticipatory workspace for macOS
Your next step, already set up.
vajco ai observes your work on-device, predicts your next action, and proposes it before you reach for it. One keypress to approve. Nothing runs without you, and nothing leaves your Mac.
⌘J to summonor it surfaces on its own
Built to a 300 ms panel budgethotkey to panel, p95
Works with the network offprediction runs on-deviceProduct demo
Watch the loop close
You push. The deploy breaks while your head is elsewhere. vajco ai notices, stages the fix, and waits for your keypress.
You push a commit
➜ lumen
Illustrative replay of the current pre-release build’s flow. UI recreated from the shipping panel; timings compressed.
The work isn’t hard.
Restarting it is.
Solo developers lose their momentum in the seams between tools: after a commit, around a deploy, on the way back from a break. The cost is never the fix itself. It is figuring out what the next step even is, then setting it up again.
- ✕You push, then tab awayThe deploy runs while you start something else. By the time it fails, your head is two contexts deep.
- ✕You rebuild the picture by handWhich repo, which branch, which error, which file. Every loop restarts with minutes of setup nobody counts.
- ✕The fix waits on you noticingDashboards do not come to you. Broken deploys sit unnoticed until you happen to look.
Anticipation, not automation
vajco ai watches your activity locally (apps, repos, branches, deploy status) and keeps a live prediction of your most likely next step. When confidence is high it surfaces one calm proposal. Press Continue and the environment is restored; press Hand off and the fix goes to Claude Code with full context.
Built around one rule:
propose, never presume
Everything below serves the same loop: see the next step early, stage it, and wait for you.
Predicts your next action
A multi-signal, on-device predictor reads recency, deploy status, and your action sequences to rank what you’ll most likely do next, tuned to the commit → deploy → fix loop.
next-action ranking
One keypress away
⌘J summons the panel, built to a 300 ms budget. High-confidence proposals also surface on their own, quietly, and are always free to ignore.
“Why this?” is evidence, not vibes
Every proposal opens into the concrete activity records that produced it: timestamps, apps, files, repos. Never a generated rationale.
Hand off to Claude Code
Approve a hand-off and vajco assembles the context payload, then shows you the exact bytes before anything is sent. Off by default.
Continue restores the environment
The apps, the browser tabs, the terminal directory, the git branch. One press puts your workspace back the way the task needs it.
Local-first, verifiably
Observation, prediction, and summaries run on-device with Apple’s on-device models. Turn the network off and vajco ai keeps working. No telemetry, no analytics.
From signal to staged next step
- 01
Observe
vajco ai logs your activity locally: app focus, repos, branches, deploy status. Structural metadata by default; file contents stay invisible unless you opt in.
on your Mac only - 02
Predict
An on-device model ranks your likely next actions from those signals. No server, no round-trip, no training on your data elsewhere.
on-device model - 03
Propose
One calm card: what it thinks you’ll do next, phrased in a sentence, with the evidence one click away. Ignoring it costs nothing.
zero-consequence - 04
You approve
Continue restores your environment. Hand off sends the task to Claude Code, after showing you exactly what leaves the device. No press, no action.
human gate
Where the minutes actually go
v1 is deliberately narrow: the commit → deploy → fix loop for solo developers. These are the moments it exists for.
The deploy broke while you moved on
You pushed twenty minutes ago and switched to another repo.
Back from lunch, context gone
Your windows are still open, but the thread in your head is not.
Waiting on a build
The deploy is still running and you are about to tab away.
Privacy & trust
A watcher you can audit
Software that observes your work has to earn the right to. vajco ai’s answer is architectural, not contractual: prediction runs on-device, the database lives in your home folder, and the app can prove what it sees.
vajco ai is pre-release, built in public by Wovol, a solo software studio. No customer logos to show you yet. Just commitments specific enough to be checked.
Content observation is off by default
Out of the box, vajco ai sees structure only: app names, filenames, repo paths, timestamps. Reading file text or error strings requires an explicit, per-level opt-in that explains exactly what becomes visible.
Secrets never persist. Ever.
Detected passwords, API keys, and tokens are stripped before anything is written to disk or sent anywhere. No setting disables this floor.
Everything it knows is inspectable
A built-in “what vajco sees” view shows every record the app holds. Downgrade a privacy level and it offers to purge the now-disallowed data. Delete everything with one action.
External send is a two-gate exception
The only data that can leave your Mac is a hand-off you opted into in Settings and approved per-use, after previewing the exact payload. Everything else works with the network off.
Stop restarting.
Start continuing.
vajco ai is heading into beta for macOS. Join the waitlist and be first in line when it opens.