A privacy-first AI workspace for deep work. Focus tracking, task capture, an accountability coach, and a personal AI brain — all running on your machine.
A real day in LockinRa, in sixty seconds: brain-dump a plan, lock in, drift once, get caught, end with the day measured.
By 2pm your prefrontal cortex is fried. The fortieth "just five minutes" wins. Discipline doesn't scale across a whole day.
Every todo app and "focus" tracker ships your tasks, screen time, and habits to someone else's server. Your work becomes their data.
Hard site-blockers can't tell research from rabbit-holes. So you whitelist everything, and the wall becomes useless by noon.
An AI that understands your task right now, nudges you back the instant you drift — and runs entirely on your machine, so the data never has to leave.
Capture, focus, get nudged, review. Every piece talks to the same local AI brain — no tabs, no context-switching, no cloud.
Reads your week and gives a short, human coach's take. Pick the voice that actually moves you.
"Four tabs of YouTube and a half-written function. Bold strategy."
"You said 90 minutes. You did 31. Phone face-down. Move."
"Rough start, but you closed two tasks. Let's get one more."
Writes your morning briefing and end-of-day wrap. Today's plan, top priorities, drift status — on-device.
Brain-dump by voice (on-device Whisper) or text. AI extracts clean tasks and deadlines.
Per-app and per-site screen time, what you watched, focus score, drift trends, and a GitHub-style streak graph. See exactly where the day leaks.
Always-on-top focus player and an ambient mixer to set the room.
Beyond the focus guardian and coach, LockinRa bundles the tools you'd otherwise juggle across a dozen tabs — and a guided welcome tour walks you through every one on first launch.
Block the rabbit-holes with strict or mild modes — hard redirect or a gentle pop. Block straight from your history in one click.
Opt-in per-app and per-site tracking, plus "what you watched" — productive vs. leisure, split by the AI, all on-device.
Markdown notes with templates, backlinks, and AI auto-titling, plus a full-window editor. Capture a thought without leaving the app.
A command palette that jumps to tasks, notes, workspaces and actions — start a session or create a task without touching the mouse.
Streak-led habit tracking, plus a PIN-gated Bad Habits tab with a private, on-device quit coach that stays yours alone.
Life goals and a week / month / year board, a weekly planner, and a prioritized inbox — the long game and today's list in one place.
Bundle the apps, sites, notes and tasks for a project, then launch the whole set-up in one click when it's time to dig in.
A semantic memory of your patterns powers drift prediction, recommended workspaces, and well-timed break nudges — opt-in, on-device.
Optional Google Calendar & Tasks sync and one-click sign-in with Google, Microsoft or GitHub — connect only what you want.
Different work, same enemy. The Guardian learns what "on task" means for your kind of day.
Stack Overflow is research; Reddit isn't. The classifier reads the actual page against your ticket, so docs stay open and rabbit-holes get called out.
Exam season with Drill Sergeant on is a different sport. Streaks, screen-time truth, and a coach that knows you said "two chapters by six."
Literature reviews are one long rabbit-hole risk. Keep the sources, lose the sidetracks — and capture citations to notes without switching apps.
Editing days need long, unbroken stretches. Set a soundscape, launch the project workspace, and let YouTube stay "reference only" — the Guardian knows the difference.
Twelve contexts a day and no manager to protect your calendar. The morning briefing front-loads what matters; the wrap tells you where the day actually went.
No accounts. No cloud. No telemetry. The models run locally, the data stays put, and you can pull the network cable and it still works.
Classification and coaching run through Ollama on your hardware — a curated catalog of 10+ models (Llama, Qwen, Gemma, Phi, Mistral, GLM) or bring your own. Swap in one click.
Voice capture is transcribed by Whisper locally. Your microphone never streams to anyone.
Tasks, notes, and activity sit encrypted at rest in the OS keychain. Your data, your keys.
No analytics SDKs, no crash pings, no "anonymous usage." There's nothing phoning home to opt out of.
The watching is a cheap rule engine. The heavy AI only wakes when there's a real question to answer — and goes back to sleep when you step away.
while monitoring. Classification is a three-stage cascade — hard rules first, the LLM only for the genuinely ambiguous.
Models load when needed and release their RAM after idle — 15 minutes while you work, faster when you're away.
in the background. No sync loops, no telemetry pings, no update nags — the network stays quiet unless you opt into an integration.
Steps down when you're away and fully unloads models on long idle, so an open LockinRa never taxes an unattended laptop.
Dump the chaos by voice or text. AI turns it into clean tasks with deadlines.
Start a focus session. Pick the task, set the soundscape, go deep.
Drift, and the Guardian catches it — with grace timers, not nag spam.
End the day with a wrap and a coach's take. Tomorrow starts planned.
Same signals, three very different deliveries. Switch any time.
"You opened the doc, stared at it, and alt-tabbed to Reddit. A masterclass in starting. Now finish a sentence."
"No negotiation. The block is 50 minutes. You've done 12. Phone in the drawer. Eyes on the task. Begin."
"Hey — slow morning, that's okay. You still shipped two things. Let's protect the next hour and call it a win."
A task manager, a blocker, a tracker, and an AI planner — each solves a slice. LockinRa is the whole loop, and it's the only one that never sees your data.
Based on each product's public positioning, July 2026. They're good tools — they're just a different shape: cloud-first slices of the loop, where LockinRa is the local-first whole.
LockinRa keeps your tasks, notes, calendar, focus sessions and screen-time history in one desktop app, with a local AI model that reads them and helps you plan the day, catch yourself drifting, and see where the hours actually went.
It works offline and needs no account. Voice transcription and AI run on your device by default — integrations like Google, Microsoft, GitHub and Slack are opt-in, and off until you turn them on.
It's free while in beta, on Windows, macOS and Linux at v1.7.0. It's young software and still moving fast, so if something breaks or feels wrong, tell us — that's how it gets better.
Five releases since June, shipping in the open. Here's what's landed and what's next.
No card, no account, no trial clock. Download it and lock in.
When beta ends, everything on this page stays free for beta users. Pro adds the heavier machinery — paid once, owned forever.
Yes. There are no accounts, no servers, and no telemetry. Your tasks, notes, and screen activity stay on your machine, encrypted at rest in the OS keychain. The AI models run locally — nothing is transmitted for us to even receive.
No. Once the local models are installed, focus tracking, capture, coaching, and the copilot all work fully offline. Internet is only used if you opt into cloud integrations like Google Calendar or an optional API fallback.
Yes. LockinRa includes a built-in site blocker with strict and mild modes, plus the AI focus guardian that classifies each site and app against your current task and nudges you back when you drift. You can block distracting sites straight from your history in one click.
Local LLMs through Ollama — a curated catalog of 10+ models including Llama, Qwen, Gemma, Phi, Mistral and GLM — plus Whisper for voice. You can also bring your own Ollama model. Switch models, see RAM recommendations and accuracy stats, and install or delete them from inside the app. No terminal required.
Yes — all three ship from the same codebase at v1.7.0. Windows comes as an installer, an .msi for managed deployment, and on the Microsoft Store. macOS has a .dmg for Apple silicon and Intel — not notarized yet, so Gatekeeper asks once on first open. Linux has an AppImage, a .deb and an .rpm; everything works except desktop-app distraction detection under Wayland, where no protocol exists for reading window titles (the browser extension still covers the web, and X11 is unaffected). Every file, with checksums and the honest per-platform limits, is on the downloads page. An Android companion and web dashboard are further out.
Windows 10/11 (64-bit), macOS 11 Big Sur or newer on Apple silicon or Intel, or Linux x86-64 with glibc 2.35+ (Ubuntu 22.04, Fedora 36 and newer). 8GB RAM for small models, 16GB recommended for the larger ones. A discrete GPU helps inference but isn't required — LockinRa picks a model that fits your hardware and tells you the trade-offs up front.
No. The always-on part is a lightweight rule engine — under 1% CPU. The LLM only wakes for genuinely ambiguous calls, and models release their RAM after you go idle. Monitoring steps down when you're away from the keyboard, so an open LockinRa doesn't tax a laptop on battery.
No. Every model in the catalog runs on CPU; a GPU just makes responses snappier. On setup, LockinRa reads your hardware, recommends a model that fits your RAM, and shows the speed/accuracy trade-offs before you commit.
The browser extension covers Chrome and Chromium-based browsers like Edge and Brave (Firefox isn't supported yet). Outside the browser, the desktop app watches the active window directly — VS Code, games, design tools, anything — no plugin needed.
Around 90% in practice. Classification is a three-stage cascade: instant hard rules, then semantic matching against your current task, and the local LLM only for the genuinely ambiguous. When it does misfire, one click on "Wrong guess" corrects it — and it learns your patterns from there.