Back to the lab

MemryLab

Shipped

A searchable timeline of how your thinking changed.

Point it at your journals, chat exports, notes and archives. It builds a visual timeline of how your beliefs and interests moved over years. Everything stays on the machine.

  • Rust
  • Tauri 2
  • React 19
  • SQLite FTS5
  • D3
  • Ollama
Obsidian, Notion, Evernote, social archives and messaging exports, behind one detection layer.
import formats
The whole app, including the search index, with no vector database.
installer
Nothing leaves the machine. No cloud dependency.
telemetry calls
The problem

Your record of how you used to think is scattered across a dozen exports in a dozen formats, and the only tools that would index it want to upload the lot to someone else's server.

The approach

A native desktop app under a strict hexagonal architecture, so the domain has no knowledge of which database, which LLM provider, or which UI is attached. Nine port interfaces, seven SQLite-backed stores, and more than thirty source adapters behind one detection layer.

The part that was actually hard

Semantic search with no vector database. Cosine similarity runs inside SQLite alongside BM25 keyword search, and the two are fused with reciprocal rank fusion, which keeps the whole index in one file and the whole app under a five megabyte installer.

Where it landed

Shipped with a Windows installer. Eight-stage analysis pipeline covering themes, sentiment, beliefs, entities, contradictions and narrative evolution, ten interactive views, and a RAG chat interface that cites its sources. Zero telemetry, zero cloud dependency, MIT licensed.

The rest of the lab