01
Local-first editing
Markdown with frontmatter, full-text BM25 search, knowledge-graph view, calendar timeline, and Kanban. Works fully offline. No localhost server. No mandatory account.
← All projects
Desktop appPersonalised AI for Everyday Life
Memex
Native desktop app for Windows and macOS — browse, edit, and search Markdown knowledge workspaces offline. AI summarisation, spaced repetition, voice dictation, and cloud sync built in. v0.2.0 now shipping.
Relationship
Memex reads and writes the same .memex/ tree shape that the claude-memex Claude Code plugin produces. Existing workspaces open without migration. Anthril is independent of Anthropic; the plugin and the desktop app are interoperable but separately maintained.
Features
What it does.
02
AI summarisation
Cmd+Shift+I to summarise the current page. Auto-summarise on save. Bulk summarise a workspace. Provider switching between Ollama, Anthropic, and OpenAI — tokens stored in the OS keychain.
03
Spaced repetition
SM-2 scheduling with flashcards defined in standard callout syntax. State tracked per workspace. Review sessions run inside the app.
04
Voice dictation
Whisper.cpp subprocess — audio never leaves the device. One-time model download (~466MB). Transcription runs locally on macOS and Windows.
05
Page history and diff
Git-powered with gix fallback. Side-by-side CodeMirror merge view or line-diff fallback. No manual commits required — history is automatic.
06
Anthril Cloud sync
Optional. Multiplayer cursors via Yjs CRDT. Public sharing with optional password, expiry, and backlinks. Private by default — local storage is the baseline.
Get started
Get started.
Download the desktop app, point it at any folder containing a .memex/ tree, and start editing offline. Anthril Cloud sync is a separate, opt-in step.
Memex is the personalised-AI frontier’s working hypothesis. It is a local knowledge base that lives on your device, runs on your hardware, and grows more useful with every week of interaction.
The v0.2.0 release ships 13 new features: AI summarisation with provider switching, spaced repetition (SM-2), voice dictation via Whisper.cpp, git-backed page history, workspace-wide find-and-replace with wikilink rewriting, Kanban view, Mermaid editor, calendar and timeline views, PDF/EPUB import, smart autolinker, public sharing, and multiplayer cursors.
How it relates to Aurora
Memex implements Aurora-derived memory patterns alongside existing LLM providers — multi-tier local knowledge graphs, event-shaped read APIs, and consolidation cycles that compress knowledge over time. It is the primary vehicle for validating Aurora’s H3.1 (longitudinal value) and H3.2 (privacy-first parity) hypotheses in production with real cultivants.
The AI never operates on your data without your hardware. Context stays local. You own the weights your choices produce.
Deferred to v0.3
Web app (cloud-only static site), Linux build, Git-backed storage as a first-class backend, and S3 Enterprise storage are all designed in the architecture docs and deferred.