Voice in. Knowledge out.
Spoken thought, kept as structured knowledge.
VoiceMemory Engine records or ingests your voice, transcribes it on your own hardware, and turns it into searchable text assets — summaries, tags, and action items drawn out automatically. Ideas said out loud stop evaporating.
Self-hosted & open source (MIT). Audio never has to leave your own infrastructure.
How it works
From a moment of speech to a searchable asset — in four steps.
The whole pipeline runs on infrastructure you control: transcription, enhancement, and storage are all local by design.
Capture
Record in the browser or upload audio files. An installable PWA lets you capture on mobile — including offline, syncing when you're back.
Transcribe
High-accuracy speech-to-text via faster-whisper, running on your own hardware, with GPU acceleration and multilingual support (Chinese, English, and more).
Enhance
A local LLM distills each note: a summary, topic tags, extracted to-dos, and flagged risks — structure added without a cloud round-trip.
Find & use
Everything lands in structured storage with full-text and semantic search. Query by keyword or by meaning; retrieve segments, not just files.
What's inside
A complete record → transcribe → enhance → search chain.
Private by architecture
ASR (faster-whisper) and LLM enhancement (via Ollama) both run locally. Your voice notes — often your rawest thinking — stay on machines you own.
Two kinds of search
PostgreSQL full-text search for exact recall, plus semantic search (sentence embeddings + FAISS) for "I said something like this once…" moments.
Automatic structure
Summaries, tags, to-do extraction, and risk flags are generated per note, and transcripts are stored as timestamped segments — not opaque blobs.
API-first
A documented REST API (upload, record, list, detail, audio playback, status) with API-key and JWT auth, so other products can use it as their voice inlet.
Realtime pipeline
WebSocket updates and streaming transcription with heartbeat detection — watch a note take shape as you speak, not minutes later.
Built to operate
Docker Compose deployment, async workers (Celery + Redis), rate limiting, circuit breakers, health checks, backup and recovery runbooks.
Who it's for
For people whose best thinking happens out loud.
Founders & operators
Capture ideas, tasks, and risks the moment they occur — walking, driving, between meetings — and trust they'll be findable later.
Product ecosystems
Give a whole family of applications one consistent voice-intake API instead of rebuilding recording and transcription in every product. This is m520's role inside LIGHT HOPE.
Teams that talk
Turn spoken collaboration — standups, debriefs, voice memos — into a shared, searchable knowledge base with the action items already pulled out.
Access & pricing
Open engine. No price tag to hide.
Self-hosted, MIT-licensed
VoiceMemory Engine is infrastructure, not a subscription. The engine is MIT-licensed and runs on your own hardware with Docker Compose — there is currently no paid hosted plan, and this page won't pretend otherwise.
- Self-host: deploy the full stack (FastAPI + Celery + PostgreSQL + Redis) yourself, free under the MIT license.
- Ecosystem integration: LIGHT HOPE projects can adopt the voice-intake API — contact us to get wired in.
- Hosted access: interested in a managed instance? Write to us; we'll talk about what you actually need.
Contact
Tell us what you'd use a voice memory for.
Deployment questions, API integration, or a pilot of a managed instance — one email reaches the people who build it.
info@x1000.ai