Skip to main content

On-Device AI Features

Fastxt includes on-device AI capabilities that help you organize, search, and understand your notes — processed locally on your device, not sent to the cloud.

Features​

Smart Tagging​

Generate relevant tags from note content. The AI prefers your existing tag vocabulary so suggestions stay consistent.

  • Desktop: Ollama or any OpenAI-compatible local server (llama.cpp, Foundry Local, LM Studio)
  • iOS/iPadOS: Apple Foundation Models on iOS 26+ where available; Natural Language heuristics elsewhere
  • Android: keyword and extractive heuristics

Summarization​

Concise summaries of long notes (desktop, iOS, Android).

Find notes by meaning, not just keywords. On desktop, semantic search combines the FTS5 full-text index with vector embeddings, fused with reciprocal rank scoring. Works for CJK text too (trigram index).

AI Organization​

Group notes into categories. Existing category names are reused (so renames stick), and you can rename or dismiss categories afterwards.

Setup​

Desktop (macOS, Windows, Linux)​

Fastxt desktop needs a local LLM backend; Ollama is the default.

  1. Install Ollama:

    curl -fsSL https://ollama.com/install.sh | sh

  2. Pull the models (a chat model for tags/summaries, an embedding model for semantic search):

    ollama pull llama3.2
    ollama pull nomic-embed-text

  3. Configure Fastxt:

    • Open Fastxt Desktop → AI Settings
    • Backend: ollama (or llamacpp, foundry-local, openai for any OpenAI-compatible server)
    • Endpoint (default http://localhost:11434 for Ollama), chat model, embedding model
    • Save, then click "Test Connection" — it probes the endpoint you configured and lists any missing models

iOS / macOS​

On iPhone and iPad, Fastxt uses Apple Foundation Models on iOS 26+ when the system model is available, and falls back to Natural Language heuristics on other devices. No setup needed.

Android​

Tag suggestions and summaries use on-device heuristics; no setup needed.

Sync​

Devices sync peer-to-peer: one device runs a server and shows a pairing code (QR or text); the other device scans or pastes it. The channel is TLS-encrypted with the server's certificate pinned from the code, and a session token authorizes every request. AI tags, summaries, categories and embeddings sync along with the notes (embeddings only between devices using the same embedding model and note version).