Skip to main content

How ɳClaw works

Infinite memory, auto-topic detection, Rust-powered retrieval, and 13 plugins — composing into one conversational AI that knows your life.

🧠

Infinite memory

Every conversation, remembered forever

Most AI assistants forget everything when you close a chat window. ɳClaw stores every conversation in a Postgres knowledge graph powered by pgvector and the ltree extension. When you ask about something you discussed three months ago, ɳClaw finds it via semantic vector search and pulls the relevant context into the prompt window.

  • Every message stored with embedding in your own Postgres
  • Full-text and vector search across your entire conversation history
  • Context window automatically assembled from relevant past conversations
  • No external cloud storage — your data stays on your server

Technical details

Storage
Postgres 16 + pgvector 0.7
Hierarchy
ltree topic paths
Search
MeiliSearch + vector cosine similarity
Context assembly
Rust (libnclaw) — chunked, scored, deduped
🌿

Auto-topic detection

No "New Chat" button, ever

After 3 messages in a conversation, ɳClaw classifies the topic using the AI plugin. The topic becomes an ltree path in your knowledge graph. Related conversations are automatically linked via pgvector similarity. The sidebar shows your life organized by subject, not by timestamp — and it builds itself without any manual tagging.

  • Topic auto-assigned after 3 messages — no manual categorization
  • Topics stored as ltree paths: /work/project/feature, /health/fitness
  • Related topics linked automatically when vector similarity exceeds threshold
  • Entire knowledge tree browsable in sidebar by subject, not date

Technical details

Trigger
Message 3 in any conversation
Classification
AI plugin (configurable model)
Storage
ltree path in np_topics table
Linking
pgvector cosine similarity > 0.85 threshold
⚙️

Rust core

Native speed for embedding and retrieval

ɳClaw's embedding pipeline and knowledge graph retrieval are implemented in a Rust library called libnclaw. This Rust core handles chunking, embedding, vector scoring, and ltree path resolution. It ships as a native library in the macOS and Linux desktop builds via Tauri, and as an FFI module in the mobile builds. No Python overhead on critical paths.

  • libnclaw Rust library handles all embedding and retrieval on critical paths
  • Tauri desktop apps on macOS, Windows, and Linux include the native binary
  • Mobile apps use Rust via FFI — no JavaScript bottleneck for vector ops
  • Chunking strategy is topic-aware — splits at sentence and paragraph boundaries

Technical details

Library
libnclaw (MIT, nself-org/nclaw)
Desktop integration
Tauri 2 — native sidecar
Mobile integration
FFI — React Native module
Embedding model
Configurable — local GGUF or OpenAI-compatible API
🔌

Plugin ecosystem

13 plugins compose into a full personal assistant

The ɳClaw bundle ships 13 nSelf plugins. Each plugin extends ɳClaw's conversational surface with a new capability. The mux plugin routes structured tool calls to the right plugin. All plugins communicate over the same Postgres bus, so context and memory flow across capabilities automatically.

  • ai: reasoning, multi-turn, system prompt configuration
  • claw + claw-web: the ɳClaw memory and knowledge graph engine
  • mux: intent routing — the conversational API gateway
  • google: Gmail read/write, Calendar events, Drive search
  • voice: speech-to-text and text-to-speech via Whisper-compatible API
  • browser: headless Chrome control for web research and form filling
  • notify: push notifications to mobile via FCM

Technical details

Bundle size
13 plugins
Router
mux plugin — intent-based tool routing
Licensing
$0.99/mo — ɳClaw bundle or ɳSelf+
Open source
Core plugins MIT; pro plugins license-gated
🔒

Full data sovereignty

Your data never leaves your server

ɳClaw runs on your own VPS or local machine. Postgres, pgvector, Redis, Hasura, and the embedding pipeline all run on infrastructure you control. The MIT client apps connect to your own backend — there is no ɳSelf-operated server in the data path for any conversation content.

  • All conversation data stored in your own Postgres instance
  • Embedding API calls go to your chosen provider — not through nSelf
  • nSelf CLI manages the backend locally — no remote shell access required
  • Open source client apps — audit the code that talks to your server

Technical details

Backend
Self-hosted via nSelf CLI
Hosting
Any Linux VPS — Hetzner, DigitalOcean, your own hardware
AI model
Your choice — OpenAI API, local GGUF, any OpenAI-compatible endpoint
Sync
Optional — all local by default

Ready to try ɳClaw?

MIT licensed, free to download. Add the $0.99/mo ɳClaw bundle to enable all 13 backend plugins and the full infinite memory stack.