· portable llm wiki
The open protocol

Your LLM doesn't
know you.

One markdown folder. Every model — Claude, ChatGPT, Cursor, Gemini — knows you.

Stop re-explaining yourself every conversation. The portable LLM wiki is a vendor-neutral personal context store: your decisions, relationships, and operating principles in plain markdown, in your git, queryable over HTTP or MCP. Watch it work below — the answer is streaming live from the wiki.

POST /wiki/chat/stream
Question
Answer
retrieving from the wiki…
Recommended
free tier
Deploy to Render →
Your own private wiki, live in ~60 seconds. Render auto-generates a secure owner token; bring your own markdown.
one-click Blueprint → render.yaml
Read the spec
MIT
Star on GitHub →
One repo. Backend, frontend, MCP server, demo wiki. Fork the protocol or run it as-is.
github.com/professorpalmer/portable-llm-wiki

Three operations.
That's the whole protocol.

see the spec →
Anything in.
Ingest
Slack thread, screenshot, voice memo, PDF, raw note. An LLM digests it, extracts entities, cross-references with [[wikilinks]], and never loses provenance back to the source.
POST /owner/ingest
Cited answers out.
Query
Natural-language Q&A grounded in your markdown. Graph-aware retrieval expands by wikilinks. Every answer carries the pages it cited — like the box above.
POST /wiki/chat/stream
Self-maintaining.
Lint
Four Puppetmaster agents run in parallel: contradictions, stale claims, missing pages, public-tier leaks. They surface issues for review — never silently rewrite.
POST /owner/lint/swarm

Plug it into any LLM.

The wiki is vendor-neutral by design. Cursor, Claude, ChatGPT, Gemini, a local Llama — anything that can fetch a URL or speak MCP gets your context for free.

connection guide →
HTTP — works with anything
Paste this URL into any chat that fetches URLs.
https://your-wiki.example.com/api/backend/.well-known/llm-wiki.json
MCP — typed tool calls
For Cursor, Claude Desktop, any MCP client.
full instructions →
{
  "mcpServers": {
    "portable-llm-wiki": {
      "command": "npx",
      "args": ["-y", "portable-llm-wiki-mcp"],
      "env": { "WIKI_BASE_URL": "https://your-wiki.example.com" }
    }
  }
}

Host your own.

three paths, pick yours

The whole protocol fits in one repo. No accounts, no hosted plan. Your markdown, your token, your URL.

Recommended
Cloud
One-click on Render's free tier.
render.yaml → Blueprint
Render provisions backend + Postgres-grade disk, auto-generates an owner token, and exposes a public URL. Pair with Vercel for the frontend.
Deploy →
For tinkerers
Local
One command, no Docker.
npx create-portable-llm-wiki
Spins up the backend, frontend, and a fresh wiki folder on your laptop. Hot-reload, demo data optional. Best way to fork the protocol.
On GitHub →
For self-hosters
Docker
Compose up on your own box.
docker compose up
A pinned image of the backend + frontend, ready for a homelab, Tailscale VPS, or whatever you already host. Mount your wiki folder; you own the data.
Compose file →

About this demo

you're seeing the wiki-demo/ seed

This deployed instance is seeded with Avery Chen, a fictional founding engineer at Strand Bio — a synthetic-biology startup. Twenty-five pages across her entities, operating principles, career decisions (leaving grad school, joining Strand, postponing the Series A), saved deliberations, and source digests. Avery is here so the wiki has something real to talk about; when you host your own, you replace the folder with your own markdown.

browse the demo pages →see the graph →open the full chat →