lore
Source
Download
Talk to Claude Code.
Hold fn, say it the way it comes out, release. All of it shows up at your cursor. Claude Code, Cursor, anywhere.
Claude Code v2.1.283
Opus 5.5 · Claude Max
~/projects/api
❯ Okay, so the login test is flaky again. I think it’s the retry. It only fails when the token expires mid-request, so can you look at the refresh logic and add a test that reproduces it first?
? for shortcuts
The film
One key.
If macOS already uses fn for something else, onboarding offers Right Option or a key you record yourself.
-
Hold fn.
The bubble appears and its timer starts.
-
Say it the way it comes out.
Any language, or several. Russian, English, half of each in one sentence; no language switch to set.
-
Release.
The text is pasted at your cursor. lore never presses Enter for you.
esc cancels without pasting. The dictation still goes to History.
fn+Space
No time limit.
A dictation runs as long as you keep talking. Lock it with Space and put your hands down.
Ramble.Correct yourself mid-sentence.Keep going.
fn+S
Take a screenshot while you speak.
Anything you copy while you talk joins the text at the clause you were saying: a screenshot, a link, a snippet, or a file if you switch that on.
Terminals and editors get the path. Web composers get the file.
❯ Okay, so the login test is flaky again. I think it’s the retry. It only fails when the token expires mid-request, so can you look at the refresh logic and add a test that reproduces it first?
⏺ Bash(npm test -- auth)
⎿ ✗ refreshes an expired token
POST /auth/refresh 200
GET /me 401
❯ so the refresh itself works. It’s the retry, it fires before the new token is saved <screenshot>…/5F3A…-1.png</screenshot> look at the second request, it still carries the old token.
? for shortcuts
Clean up only when you ask.
Raw words paste by default.
Both use your own OpenAI key.
fn+V
Clean up on paste.
- said
- put the cache in
, uh, redis — no,postgres - pasted
- Put the cache in Postgres.
Self-corrections resolved, no paraphrasing.
fn+T
Translate to English.
- said
- давай вынесем кэш в отдельный сервис
- pasted
- Let’s move the cache into a separate service.
Every dictation kept.
Every dictation stays in History with its text and duration, editable, with the audio for the last 500. Stats shows your words, your time, and a heatmap by day.
The audio is written to disk from the first second. If the app is killed mid-sentence, the dictation is in History at next launch, with a Retry.
since apr 30, 2026
Meetings, too.
Record a meeting from the mic and system audio, and get a transcript that separates you from the other side, with notes made on your Mac.
Recording a conversation is on you: many places require everyone’s consent first, and lore leaves that, and the law, in your hands.
- You
Can we ship the export on Friday, or is the migration still in the way?
- Them
The migration is done. What’s left is the review.
- You
Then let’s review it today and ship on Friday.
On your Mac.
Transcription runs on your Mac. Text leaves it only when you ask.
- Dictation cleanup and translation
- OpenAI, with your own key
- Live meeting cleanup and Ask Lore
- OpenAI, with your own key
- A problem report
- Our server, only when you press Send, exactly as previewed
- Update check
- At launch and every six hours. Settings has a switch.
- Speech files
- Downloaded once from Hugging Face
No account, no analytics, no cookies.
- Your data
- ~/Library/Application Support/Lore
- Speech files
- ~/Library/Application Support/FluidAudio
What it does not do.
- Mac only. macOS 15 or later, Apple silicon.
- Text appears on release, not while you speak.
- Names and jargon get guessed. There is no custom dictionary, on purpose; the agent on the other end usually knows what you meant.
- Cleanup and translation need an OpenAI key and go to OpenAI. Without a key, raw text pastes.
- Live translation is into English only. Other languages are available afterwards from the History row.
- Not tested with VoiceOver.
Get lore.
- Drag lore to Applications and open it.
- Allow three permissions: Microphone, Accessibility, Input Monitoring.
- lore downloads its speech files once, about 930 MB.
- Hold fn.
Open source.
MIT license. To build it yourself:
$ git clone https://github.com/atarazevich/lore$ cd lore/app && ./build.sh
Needs a full Xcode install and an Apple Development certificate. The script signs the bundle and refuses to build unsigned.