Frequently asked questions
Everything you need to know about ɳClaw — the self-hosted AI personal assistant that never forgets.
- What makes ɳClaw different from other AI assistants?
- ɳClaw has infinite memory. Every conversation is stored in a Postgres knowledge graph powered by pgvector and ltree. When you ask a question, ɳClaw searches your entire conversation history via vector similarity and pulls relevant context into the prompt window. Most AI assistants forget everything when you close the chat. ɳClaw never does.
- Is ɳClaw free?
- The client apps (desktop + eventually mobile) are MIT licensed and free forever. You can self-host the backend for free using the nSelf CLI. The ɳClaw bundle ($0.99/mo) unlocks 13 backend plugins including the AI reasoning engine, email, calendar, voice, and browser control. Without the bundle the backend runs but the AI plugin is not active.
- How does self-hosting work?
- You clone the nclaw repo, install the nSelf CLI, set your ɳClaw bundle license key, and run "nself build && nself start". The CLI provisions Postgres with pgvector, Redis, Hasura GraphQL, and all 13 plugins in Docker on your own server or local machine. The whole stack runs in one command. You need a domain pointed at your server; nSelf CLI handles TLS via Let's Encrypt automatically.
- Where is my data stored?
- All conversation data is stored in your own Postgres instance on your own server. ɳSelf does not operate any server in the data path for conversation content. AI model API calls go directly from your server to your chosen provider (OpenAI, or any OpenAI-compatible endpoint). Nothing passes through nSelf infrastructure.
- What AI models does ɳClaw support?
- ɳClaw uses the AI plugin which supports any OpenAI-compatible API endpoint. You can use OpenAI directly, a local model via Ollama or LM Studio, or any provider with a compatible API (Anthropic via proxy, Groq, Together AI, etc.). The model is configured per-plugin via your nSelf environment settings.
- How does auto-topic detection work?
- After the third message in any conversation, the AI plugin classifies the topic. The topic is stored as an ltree path in the np_topics table (e.g. /work/project/feature, /health/fitness/running). ɳClaw then computes pgvector cosine similarity between the new topic and existing topics. If similarity exceeds the threshold, the conversation is linked to related topics automatically. The sidebar in the app shows your full topic tree.
- Is there a mobile app?
- The React Native mobile app (iOS + Android) is in active development and will ship in a future release. The current v1.x release includes desktop apps only (macOS, Windows, Linux) via Tauri. Follow the GitHub repo for release updates. The desktop apps are fully functional today.
- What is libnclaw and why Rust?
- libnclaw is the Rust library that powers ɳClaw's embedding pipeline and knowledge graph retrieval. It handles text chunking, embedding model calls, vector scoring, and ltree path resolution. Using Rust for these paths eliminates Python overhead on the critical operations. libnclaw ships as a native sidecar in the Tauri desktop builds and as an FFI module for mobile. It is MIT licensed and part of the nclaw monorepo.
- Can I use ɳClaw without the ɳClaw bundle?
- Yes. The client app (desktop) connects to your self-hosted backend. Without the bundle the backend runs but the AI, claw, and claw-web plugins are not active — which means no AI responses and no memory graph. The bundle is what activates the core ɳClaw experience. At $0.99/mo it is the minimum required license to run ɳClaw as an AI assistant.
- What is included in the ɳClaw bundle?
- The ɳClaw bundle ($0.99/mo) includes: ai (reasoning, multi-turn), claw (memory engine), claw-web (web frontend sync), mux (intent routing), google (Gmail + Calendar), voice (Whisper-compatible STT/TTS), browser (headless Chrome automation), notify (push notifications), and 5 additional utility plugins. 13 plugins total. The ɳSelf+ plan ($3.99/mo) gives you all 6 bundles and every companion app.
Still have questions?
Open an issue on GitHub or check the self-hosting documentation.