What is Memory.Wiki

Personal knowledge hub for the AI era.

Memory.Wiki is a single URL that holds everything you've ever asked an AI worth keeping, and that any AI tomorrow can read as your context.

The one-line version

A captured ChatGPT answer becomes memory.wiki/<id>. Forty captures roll up into memory.wiki/hub/<you>. You paste that hub URL into Claude, ChatGPT, Cursor, or Codex and they fetch the whole thing as your personal knowledge layer. Same URL, every AI.

Why a URL

Vendor memory layers (ChatGPT memory, Claude projects, Cursor docs) all live behind walls. They don't talk to each other, you can't share them, and you can't paste them into the other tool tomorrow.

A public URL is the one format every AI already knows how to read. No SDK, no plugin, no MCP server required. The URL is the contract; Memory.Wiki makes the URL.

Three primitives, one shape

  • Document at memory.wiki/<id> is one captured answer, paper, or transcript.
  • Bundle at memory.wiki/b/<id> is a curated grouping of docs around a topic, with its own URL.
  • Hub at memory.wiki/hub/<you> is your whole knowledge layer as a single deployable URL.

Same URL primitive, three scopes. Each one is plain markdown when an AI fetches it.

Three operations

  • Capture: paste a share URL, drop a file, or run /memory.wiki capture from any coding agent.
  • Organize: captures roll up into your hub; bundles group by topic; auto-synthesis merges new captures into existing wiki pages.
  • Deploy: paste any of the three URLs into any AI and it loads the underlying markdown as context. Or hit /api/hub/<slug>/recall for question-targeted retrieval.

What's not on the surface

Underneath, every doc, chunk, and bundle is automatically embedded (OpenAI text-embedding-3-small, idempotent on save). Hybrid BM25 + vector RRF retrieval. Hub-wide semantic graph. Cross-hub citation rollup. The pipeline rivals what backend-only memory systems give you, but the surface stays human-shaped.

The cross-AI claim is independently verifiable. MWBench is the open eval that measures whether one URL actually works across Claude, OpenAI, and Gemini: 100% accuracy, including on content the AIs have never seen during training, with every answer audited against a literal corpus quote.

For the architecture, see How Memory.Wiki works and How Memory.Wiki Memory works.