amem Clipper

About

amem Clipper turns your browsing into agent memory.

Press one button and the page you're on becomes a "memory ball" — a node in your local knowledge wiki, complete with title, source URL, host favicon, an AI-generated 2-sentence summary, and a stable node_id so the same URL never duplicates.

Click any ball later to jump back to the source content. Open the wiki folder in Obsidian and your captures appear as a real knowledge graph.

WHAT IT DOES

• One-click capture — the Snapshot button extracts the page's text and routes it through your local amem-librarian daemon for summarisation.
• Auto-summarisation — uses your existing Claude Code subscription via `claude -p` (no extra cost, no extra account). Falls back to a local Ollama model when offline.
• Local wiki — every ball persists as a markdown file at ~/.amem/wiki/<node_id>.md with full YAML frontmatter and an Obsidian-compatible Connections section. Yours to grep, git, or render.
• Deduplication — the same URL captured twice updates the existing node and appends to its captured_at history. No duplicate cards.
• Click-through — ball cards open the source URL. arxiv balls open the PDF inline; X, Anthropic, and other sites open the original page.
• No telemetry, no cloud — the daemon runs on your machine; the wiki lives on your disk; nothing leaves localhost except page fetches you initiated yourself.

WHY IT EXISTS

If you use Claude Code, Cursor, or Cline daily, you've noticed the gap: your AI knows the public web but not the things you've actually read. amem fixes that. Your reading becomes a citeable corpus your agent can search via MCP.

REQUIRES

• macOS (Linux/Windows in roadmap)
• The amem-librarian CLI running locally — github.com/yiidtw/amem-librarian
• Either a Claude Code login (default) or a local Ollama install (fallback)