How it works

The whole pipeline runs on your computer.

One background process on your Mac pulls in your data, understands it with local models, and serves it to apps and AI tools through a local API.

YOUR COMPUTERMessagesEmailNotesVoiceApp dataTapestri appsYour appsClaude, ChatGPT…01Connectorsread-only02Encrypted storeone schema03Local modelsextract · transcribe04Indexsemantic + keyword05API + MCPscoped tokensNothing in this pipeline needs a network connection. Sharing is a separate, opt-in step.Messages · Email · Notes · Voice · App dataYOUR COMPUTERConnectors01 · read-onlyEncrypted store02 · one schemaLocal models03 · extract · transcribeIndex04 · semantic + keywordAPI + MCP05 · scoped tokensTapestri appsYour appsAI toolsNo network needed. Sharing is opt-in.

  1. Capture

    Read-only connectors pull from local app databases, official APIs and watched folders, then sync incrementally.

    read-only · incremental

  2. Normalize

    Everything maps to one schema, duplicates merge, and each record links back to its original.

    one schema · provenance on every record

  3. Extract

    On-device models transcribe audio, read screenshots, and find people, places and events, merging each person's numbers, emails and handles.

    person "Maya Chen" · 214 mentions · 6 sources

  4. Index

    Search combines meaning, exact keywords, and filters for time, people and source.

    semantic + keyword + filters

  5. Serve

    A local API and MCP server give each app or AI its own scoped token, and every read is logged.

    per-client scopes · access log