Drop in your sources. An AI compiles them into a wiki that compounds.
You curate articles, papers and PDFs; an AI agent reads them, writes interlinked Markdown pages, and keeps the index and log current. Cursor, Claude and ChatGPT read and write the same wiki through MCP, so knowledge stops evaporating in chat threads.
Compiled, not retrieved
Not notes in the cloud for an AI to fish chunks from: every source is read, summarised, cross-referenced and checked for contradictions as it arrives. Every question stands on knowledge that is already organised.
Any agent can read and write
Built-in MCP server: Cursor and Claude Code paste a token; Claude.ai and ChatGPT sign in with OAuth. On the web, one click ingests with your own API key.
Built for researchers
Bibliography filled in automatically (DOI, Crossref, arXiv, PubMed), BibTeX and CSL-JSON in and out, [@citekey] citations with an automatic reference list, Zotero sync.
How it works
1. Add sources
Paste a URL, upload a PDF or Word file, paste text, import a .bib, or connect Zotero. Sources land in raw/ as pending.
2. Let the agent ingest
One click on the web, or paste the prompt into Cursor. The agent reads the rules in schema/, then writes the summary page and updates related pages, index.md and log.md. Discuss first if you want to steer.
3. Ask, browse, lint
The chat panel answers with citations and files good answers as wiki pages; Lint finds orphans, broken links and contradictions. Graph, table view and version history are always there.
The agents you already use read and write this wiki
Cursor and Claude Code paste a token; Claude.ai and ChatGPT sign in with OAuth. The same six tools everywhere: read rules, search, read, create, update, list.
Knowledge grows into a graph
Every [[link]] and [@citation] on a page is an edge. The more sources, the clearer the hub pages; the timeline replays how the wiki grew.
Who it is for
Researchers
Weeks or months of deep reading that accumulate into a cited review draft.
Knowledge workers
Meeting notes, decisions and project documents get organised on arrival, so nobody digs through chat history.
Heavy Cursor and Claude users
Persistent memory across projects and conversations, in a Markdown base you can open and export any time.
Plans
No card. Everything, unlimited agent runs; the first 10 runs need no API key at all.
200 notes, 20 agent runs a month, bring your own API key.
10,000 notes, 1 GB, unlimited agent runs, multiple tokens, 90-day version history.
Model costs are not included: you use your own key, typically a few cents to a few dollars a month; Cursor users need no key. Checkout by Paddle; cancel any time.
Open-source core, your data in your hands
The code is released under AGPL-3.0 and self-hosts with docker compose; the hosted version is run by us. Export the whole wiki any time as an Obsidian-compatible Markdown zip, and the bibliography as .bib. Interface in Traditional Chinese and English; write in any language.