Ask your own day

what did I agree to this morning?

Recalled

You took the boundary fix, and said you'd have it behind a flag by end of day.

Standup · desktop microphonePromise
Follow-up · desktop microphoneDecision

Open-source audio memory. NeoRecall records from the devices you already use, runs on the inference providers you configure, and deletes each recording once its transcript is proven durable.

Open source · Web · macOS · Windows · wearable sources

Most of a day vanishes because nobody had time to write it down. NeoRecall is built the other way around: it listens with consent, uses inference services you choose, and keeps the useful parts — not the recording.

01

It keeps up

Durable chunks survive weak connections and interrupted days. Every source can record offline and resume safely when the server returns.

02

It understands

Configured transcription and language-model services, combined with local search, turn raw conversation into names, decisions, promises, and moments.

03

It lets go

A terminal receipt proves the transcript is durable and the server copy is deleted before any client is allowed to release its audio.

From sound to recall

Scroll. Your day is becoming searchable.

Capture · live transcript
10:14 · desktop microphone
Let’s use the smaller enclosure. I’ll send the revised dimensions after lunch.
Capture Transcription API Speakers
Promise detected · Revised enclosure dimensions will be sent after lunch. Linked to the hardware planning conversation.
Processing externally · audio retained until terminal receipt
Memory · connected context
episodicImp 894%

The smaller enclosure was chosen; revised dimensions are due after lunch.

Privacy · terminal receipt

Chunk 10:14:30

SHA-256 verified · idempotent upload
Finalizing
Transcript persisted durably
Speaker evidence saved
Server audio unlinked
Client release authorized
The transcript stays. The original audio does not.
Capture

Keep living. NeoRecall keeps up.

Record from web, desktop, imports, or a wearable. Independently decodable chunks queue locally, upload idempotently, and survive the ordinary failures of an ordinary day.

Understand

One conversation becomes one useful memory.

Your configured transcription and language-model endpoints work with NeoRecall's multilingual retrieval and its own database. Search exact words or ask naturally across people, facts, tasks, and decisions.

Release

Deletion is a protocol, not a promise.

The client holds every audio chunk until a terminal receipt proves transcript persistence and server-side deletion. Only then can it release its own copy.

today · recording
CAPTURE

Record what matters

Desktop can capture microphone and system audio together.

STANDBY Ready to record
Nothing waiting to processAudio is released only after a terminal receipt.
CAPTURE SOURCES
MicrophoneVoice and nearby sound✓
System audioCalls and desktop playback○
BluetoothPreferred wearable device✓
Import an audio fileWAV, MP3, M4A, and more
CAPTURE

Record what matters

Desktop can capture microphone and system audio together.

STANDBY Ready to record
Nothing waiting to processAudio is released only after a terminal receipt.
CAPTURE SOURCES
MicrophoneVoice and nearby sound✓
System audioCalls and desktop playback○
BluetoothPreferred wearable device✓
Import an audio fileWAV, MP3, M4A, and more
One memory, many sources

Record where the day actually happens.

Start on the web, keep going on a desktop, import an old recording, or connect a wearable. Every client follows the same durable capture contract, so changing devices does not create a hole in your day.

  • Web, macOS, Windows, and wearable capture
  • Microphone, optional system audio, and file imports
  • Offline queues with automatic, idempotent resume

One private memory, wherever the sound comes from —

MicrophoneSystem audioAudio importsWebmacOSWindowsWearablesDiscordMeeting botSmart panel

— normalized into one timeline, one set of speakers, and one place to ask what happened.

Yours, actually

Your memory stays yours.

NeoRecall keeps its database, search index, and durable capture state on one machine you control, and works out who is speaking there too. Transcription and memory writing always use services you configure separately, whether those are hosted APIs or endpoints on another machine you control.

  • Open source, AGPL-3.0 — inspect every layer
  • Bring any supported provider or compatible endpoint
  • SQLite memory that you can search, back up, or erase
neo@home-server — zsh
$
✓ cloned NeoLabs-Systems/NeoRecall (stable)
✓ NeoRecall CLI linked
$ neorecall install
✓ user service configured
$ neorecall setup
✓ semantic search verified
✓ control surface on http://localhost:4500/app/
Configure inference providers in Admin to begin.

Remember the day.
Not the recording.

Install NeoRecall on your own machine, connect one source, and give yourself a memory you can actually search tomorrow.