Private audio memory you host yourself.
Records your day anywhere, turns every conversation into a searchable memory,
and answers from it. On your hardware, with the models you choose.
neorecall-promo.mp4
|
Your server, your database, your search index. Desktop app, Docker or one script. |
Open source under AGPL-3.0. Any transcription and language model, local ones included. |
Audio is removed only after the transcript is durable. |
| 🖥️ Desktop app | macOS 13+ and Windows 10/11. Set up NeoRecall on this computer installs the server for you, no terminal. | Download |
| 🐳 Docker | Home servers, NAS boxes and VPS. | Guide |
| 📜 Script | Linux or macOS with Git and Node.js 20+. | Guide |
Script
bash <(curl -fsSL https://raw.githubusercontent.com/NeoLabs-Systems/NeoRecall/main/install.sh)Then open http://localhost:4500 once neorecall status reports a running service. The first account you create becomes the administrator; the app then asks for your transcription and language-model services (or an on-host model) and tests both.
| 🎙️ Records anywhere Browser, macOS and Windows with microphone and system audio. Quick capture sits in your tray. |
🧠 Memories, not transcripts Each conversation becomes one memory with people, promises, decisions and tasks, written when it ends. |
💬 Ask your day "What did I agree to this morning?" Answers cite the moments they came from. |
| 🗣️ Knows who said what Speech and voices are detected on your machine (31 MB of models). The same voice is recognized weeks later. |
♾️ Never stops Upload, transcription and conversation detection run during capture. Long recordings are read in windows and folded into one memory. |
📴 Offline first Clients buffer decodable chunks and resume idempotent uploads when the server is back. |
| 📟 Wearables and devices Omi, Plaud, Memoket and more over Bluetooth, plus NeoRecall Desk, a Raspberry Pi call recorder. |
🎧 Meetings and voice Zoom, Teams, Webex and FaceTime detection, and a Discord notetaker bot you run yourself. |
🔌 NeoAgent and MCP Connect NeoAgent, or paste the MCP URL into Claude, ChatGPT or Cursor for read-only recall. |
| 📝 Context Add highlights, notes, photos and files while recording. They attach to the right memory. |
🛡️ Consent built in A recording notice and a persistent recording state. There is no covert mode. |
👥 Multi-user Per-account memory, admin role, two-factor and security keys, encrypted backups. |
Inference: bring any OpenAI-compatible transcription and language-model service, or run Whistle (speech to text) and Gemma 2 2B (memories and Ask) on the server itself. NeoRecall bundles no model you did not choose.
| 🕒 Moments | 🧠 Memories |
|---|---|
 |
 |
| 💬 Ask | 🔎 One memory |
 |
 |
| Start | Use | Run |
|---|---|---|
| Installation | Recording | Configuration |
| Privacy and consent | NeoRecall Desk | Troubleshooting |
| Architecture | Contributing | Security |
Full docs: neolabs-systems.github.io/NeoRecall/docs
- Beta. Maintained mainly by one person. Expect breaking changes.
- Privacy. Read privacy and consent and get the consent your jurisdiction requires before recording anyone else.
- Help. Discussions for questions, Issues for bugs, SECURITY.md for vulnerabilities.
Made with ❤️ by Neo · NeoLabs Systems