PKM Capture Plugin — Frictionless Knowledge Ingestion — pyrite

Problem

PKM users (Obsidian, Anytype, Logseq) need frictionless capture: snap a photo, paste a URL, record a voice note, clip a web page. Current Pyrite workflows require CLI or web editor — too much friction for mobile/on-the-go capture. Without a low-friction capture path, Pyrite can't credibly serve the PKM audience.

Solution

A Pyrite plugin that provides capture workflows: ingest raw content (images, URLs, voice, text), classify it using AI, extract structure, and create typed KB entries automatically. The human captures; the AI files.

Entry Types

`capture` — raw ingested content before classification (ephemeral, processed into typed entries)

`clipping` — web clippings with source URL, extracted content, summary

`note` — quick text captures, voice transcriptions, brain dumps

`bookmark` — URLs with metadata, tags, summary

`reading_highlight` — extracted quotes/annotations from articles or documents

Capture Workflows

Each workflow follows the same pipeline: Ingest → Classify → Extract → Create typed entry → Link

| Input | Processing | Output | |-------|-----------|--------| | Image | OCR or vision model describes content | Typed entry (note, document reference, whiteboard capture) | | URL | Fetch, extract content, summarize | `clipping` or `bookmark` with auto-tags | | Voice note | Transcribe (Whisper or similar), summarize | `note` with transcript, extracted action items | | Pasted text | Classify, extract structure | Typed entry based on content (note, quote, reference) | | PDF/document | Extract text, chunk if needed | One or more entries with relationships | | Web clipping (browser extension or share sheet) | Extract article content, metadata | `clipping` with source chain |

AI Integration

Auto-classification: BYOK model determines entry type based on content and KB schema

Auto-tagging: Extract topics, entities, themes from captured content

Auto-linking: Suggest connections to existing KB entries based on semantic similarity

Summarization: Generate concise summaries for long-form captures

Action extraction: Pull out todos, decisions, questions from voice notes and text

Interfaces

Web UI quick-capture: Minimal form — paste/type/upload, one click to ingest. Mobile-responsive.

Claude app: Capture through conversation — "save this to my KB" with auto-classification

MCP tools: `capture_ingest`, `capture_classify`, `capture_process` for programmatic access

CLI: `pyrite capture ` for power users and scripts

Browser extension (future): Right-click → save to Pyrite

Share sheet (future): Mobile share → Pyrite capture

Prerequisites

BYOK AI integration in web UI (for classification and summarization)

Mobile-responsive web UI

Waves 1-3 shipped (platform credibility before targeting PKM audience)

Success Criteria

Capture-to-typed-entry in under 5 seconds for text/URL inputs

Auto-classification accuracy > 80% for common content types

Mobile web UI capture works smoothly on phone browsers

Obsidian vault migration: `pyrite init --from-obsidian ` creates a working KB

10+ entries captured via web/mobile in first week of use by test users

Launch Context

This is the wave 4 plugin in the launch plan. Waves 1-3 establish platform credibility with agent builders, dev teams, and researchers. Wave 4 opens the aperture to everyone who collects and organizes information. By this point, three shipping plugins prove the platform works — the PKM crowd hears "software teams, journalists, and AI agents all use this" before being asked to try it themselves.

PKM Capture Plugin — Frictionless Knowledge Ingestionbacklog_item