Lawyers & consultants
Use the built-in backend to keep audio and inference on the Mac, with a local library your organization can inspect and control.
Private AI work memory for Mac
Your private AI memory for work.
LokalBot records both sides of meetings without a bot and turns conversations plus the workday context you choose into private, searchable memory. Recall decisions with evidence, dictate, write, and automate — using on-device AI by default.
Memory you own
Meetings, notes, and context live in local files and SQLite.
On-device by default
Built-in capture, transcription, recall, and writing stay on your Mac.
Trust you can verify
Read the source, inspect the network boundary, or build it yourself.
The part nobody puts on the pricing page
Memory shouldn't have a landlord.
Every cloud notetaker keeps your working memory on servers that have to outlive funding rounds, pivots, and acquisitions. Sooner or later, every tenant gets the same letter.
Notice of memory eviction
Dear valued user,
We're incredibly proud of what we built together. Unfortunately, following our acquisition by a company you've never heard of, this service will be sunset on August 15.
Your 412 meetings — every decision, every action item, every "let's circle back to that" — will be permanently deleted. Please export a .zip you will never open.
Thank you for being part of our journey. 🚀
— every cloud notetaker, eventually
LokalBot can't send you this letter. There's no server to sunset, no account to close, no subscription to lapse. Your work memory lives in local files and SQLite on your own Mac — still readable in twenty years, with or without us.
# after a full meeting — recorded, transcribed, summarized # built-in backend, models already downloaded $ sudo lsof -i -nP | grep -i lokalbot LokalBot 4821 you 23u IPv4 TCP 127.0.0.1:17872 (LISTEN) LokalBot 4821 you 24u IPv4 TCP 127.0.0.1:17873 (LISTEN) $ # the main and embedding llama.cpp servers, $ # both bound to loopback. nothing outbound.
Prove it
Don't take our word for it.
Every AI product claims to care about privacy. LokalBot's boundary is checkable: inspect the source or run Little Snitch through a meeting. You will see model and update downloads, localhost inference, and only remote inference origins you explicitly approve.
- No account, analytics SDK, telemetry, or LokalBot AI backend
- Open source — read the network code yourself
- The bundled llama.cpp server binds to localhost only
- Non-loopback inference requires approval for the exact origin
Who it's for
Built for conversations that can't leak.
If your calls are privileged, regulated, or under NDA, where processing happens is a policy question. LokalBot's built-in mode keeps inference on the Mac; your organization still decides its consent, retention, device-security, and compliance requirements.
Therapists & clinicians
Local transcription and summaries can reduce third-party exposure. LokalBot does not itself certify HIPAA or any other regulatory compliance.
Founders & dealmakers
Board calls, fundraising, M&A, unreleased products. The notes live on the laptop they were spoken into.
Teams that banned notetakers
Local built-in inference avoids a meeting bot and hosted AI vendor. Security teams can review the source, permissions, and documented network boundary.
One memory, four moves
Remember.Recall.Write.Act.
Meeting notes are the entry point. The product is the memory that keeps working after the call.
01
Remember
Record both sides of a meeting without a bot. Add the activity, accessible text, or encrypted visual context you explicitly choose.
- Me / Them audio tracks
- Structured recaps and outcomes
- Opt-in day context
02
Recall
Ask what happened and get evidence, not a vague answer. Search by words or meaning, then open the exact meeting or retained moment behind it.
- Full-text and semantic search
- Citations and timestamp playback
- System-wide Quick Recall
03
Write
Bring that context back into any app. Dictate with ⌥ Space, or let Cotyping complete the next thought inline.
- On-device voice typing
- Inline autocomplete
- Secure fields skipped
04
Act
Turn remembered work into drafts you can inspect: follow-ups, stand-ups, work logs, exports, and approved Agent Mode sessions.
- Fixed-scope local routines
- Markdown memory exports
- Read-only CLI and MCP access
Private by construction. Built-in processing stays local. Optional remote inference and network-capable Agent Mode commands require your explicit approval.
How it works
Capture once. Recall when it matters.
-
01
Remember the moment
Start from the menu bar, or enable ask-first or automatic meeting detection. Your mic and the meeting app stay on separate tracks.
-
02
Turn it into evidence
On-device models produce a labeled transcript, structured recap, outcomes, and a searchable local index.
-
03
Bring it back into the work
Recall the decision, open its source, write the follow-up, or give a trusted tool scoped access. Approved remote inference is the disclosed exception.
Write
Put the memory back into any app.
Two ways to get words out of your head and into the window you're already in — both running on the same on-device model that writes your summaries.
Following up on our sync, the on-device build is ready to ship.
Tab to accept · Esc to dismiss
Cotyping finishes the thought inline.
Ghost text appears as you type in almost any app. Press Tab to accept a word or the whole line. With the built-in backend, the same local model that writes summaries completes your text on the Mac.
- Works system-wide, opt-in per app
- Skips password and secure fields
- Same backend as your summaries
Try it — type in the panel, then press Tab.
Listening… ⌥ Space
Dictation types what you say.
Hold ⌥ Space in any app and talk. LokalBot pauses your music, shows a live transcript in a floating pill, and pastes the words at your cursor — transcribed by the same on-device model as your meetings.
- Works in any app — paste or clipboard
- Live transcript while you speak
- Audio transcribed on-device, then deleted
Remember · Day timeline
Remember the work between meetings.
If you turn it on, LokalBot notes which app and window is in front, then builds a private timeline and one-click digest. The built-in backend answers on-device; an approved remote backend receives the context needed for your request.
- Colour-coded timeline + time-by-app totals
- One-click daily digest, generated on-device
- Ask your day in plain language
Local engines
The model is infrastructure. Your memory is the product.
Download the built-in defaults once and use them without API keys. Prefer something else? Point a task at another local model, Ollama, Apple Intelligence, or an explicitly approved OpenAI-compatible endpoint.
Compare
Same memory, different ownership.
Cloud notetakers are polished products with one structural problem: your conversations become a vendor-controlled archive. LokalBot's answer isn't a policy — it's an architecture.
| Feature | LokalBot | Cloud notetakers |
|---|---|---|
| Where the AI runs | On your Mac — Neural Engine + local llama.cpp | Vendor servers, often via third-party AI providers |
| Your audio & transcripts | Stay local with the built-in backend | Uploaded and processed in the cloud |
| Account required | None | Account, usually a subscription |
| Price | Free, open source (GPLv3) | Free tiers, then roughly $8–35 per user / month |
| Source code | Public — read every line | Closed |
Guides
Verify how your memory stays yours.
Source-backed answers to the questions that matter before you trust private meetings and work context to an AI product.
Local meeting notesHow local AI meeting notes workCapture, on-device processing, privacy boundaries, and trade-offs.
Offline transcriptionTranscribe meetings without internetWhat to download first and what keeps working offline.
Bot-free captureRecord both sides without a botHow macOS separates your mic from meeting-app audio.
Model guideChoose a local transcription modelGranite, Parakeet, Qwen3-ASR, and Whisper compared by use case.
Open sourceWhat to verify beyond the labelLicense, runtime, storage, network boundaries, and builds.
CompatibilityCheck system requirementsmacOS, Apple Silicon, memory, disk, and permissions.
Download
Give your work a memory you own.
Download LokalBot from GitHub Releases. Selected models download on first use; built-in inference can then work offline.
Download for macOS.dmg installer · all releases and notes
- Apple Silicon (M1 and later)
- macOS 15.0 or later
- Model downloads vary by your choices
Prefer to build it yourself?
$ git clone https://github.com/stevyhacker/lokalbot.git $ cd lokalbot $ xcodegen generate && open LokalBot.xcodeproj
Questions
The short answers.
Does anything leave my Mac?
Audio stays on your Mac. With the built-in backend, transcripts, summaries, screenshots, and workday context stay there too. LokalBot also connects for models, updates, optional Agent Mode setup, and any non-loopback inference origin you explicitly approve. Read the privacy policy.
Is it really free?
Yes. No account, no subscription, no telemetry. The full source is on GitHub, and you can build it yourself.
Which Macs are supported?
Apple Silicon Macs (M1 and later) running macOS 15.0 or newer. The on-device models lean on the Neural Engine, MLX, and Metal.
How does it record both sides?
Your microphone is one track, labeled Me. A Core Audio tap on the meeting app is the other, labeled Them. That split gives you speaker labels for free.
Can I use my own model?
Yes. Use the built-in llama.cpp runtime with any downloaded GGUF model, or point LokalBot at Ollama, any OpenAI-compatible server, or Apple Intelligence.
Is my screen being watched?
Only if you turn on screenshots. They are encrypted on disk and deleted after 14 days by default. Password fields and excluded apps are never captured.