VaultChat — local-first WhatsApp archive reconstruction with Phoenix LiveView + whisper.cpp

Hi everyone,

I want to share an early alpha project I’ve been building: VaultChat.

The problem

WhatsApp exports with many voice notes are hard to review. The text archive alone is incomplete — important statements often live inside audio messages. When you need to reconstruct what was said and when, manually listening through hundreds of voice notes is slow and error-prone.

What VaultChat does

It is a local-first archive reconstruction and review tool:

1. Import a WhatsApp export ZIP via browser upload

2. Parse _chat.txt across common timestamp variants

3. Transcribe voice notes with whisper.cpp — runs entirely locally

4. Rebuild a merged chronological timeline of text + audio transcripts + media

5. Search, filter by sender/type/date, export as Markdown/JSON/CSV

The key design principle: everything runs on your machine. No account, no cloud, no data leaving your machine. docker compose up and open localhost.

Stack

• Elixir / OTP

• Phoenix + LiveView

• SQLite via ecto_sqlite3

• ffmpeg for audio decoding

• whisper.cpp for local transcription

• Docker Compose

Current state

Early alpha — the main end-to-end workflow is working on real archives.

Some things I found interesting to build in Elixir:

• LiveView PubSub for real-time per-file processing progress

• ETS GenServer for progress state that survives page remounts mid-processing

• Multi-stage Docker build: whisper-cli compiled on Debian in Stage 0 and copied into the runtime image — avoids musl/glibc mismatch

• SQLite batch inserts chunked to stay under the 32766 bind variable limit

• String.replace_invalid/2 to sanitize whisper output before Jason encoding

Known limitations

• WhatsApp only for now (Telegram planned)

• Docker-first — no one-click installer yet

• Early alpha — rough edges in the UI

• No LLM features yet (planned via ollama later)

Feedback I’d appreciate

• Does the use case sound real and useful to you?

• Does Phoenix LiveView feel like a good fit for this kind of local workflow tool?

• Any thoughts on Docker-first as the packaging approach for an alpha?

• What would make you actually try it?