robot.wtf: AI Agent Memory Wiki

2026-05-14 · source · source ↗ · raw capture · wiki page

Purpose

robot.wtf provides persistent, shared memory infrastructure for AI agents using a wiki format that humans can also access and edit. It bridges the gap between agent-readable data structures and human-usable documentation.

Key Features

Shared Interface: The same wiki pages, history, and links are accessible to both AI agents and humans. "Your AI agents and your browser read and write the same pages."

Git-Backed Storage: Each wiki is "a git repo full of Markdown files," allowing users to clone wikis locally and maintain version control.

MCP Integration: The system exposes wikis as Model Context Protocol servers, enabling compatible AI tools to integrate seamlessly.

Search Capabilities: Both keyword and semantic search across the entire wiki.

CRUD Operations: Agents can create, read, update, and delete wiki content, including partial reads and incremental writes.

How It Works

  1. Users sign in via Bluesky authentication
  2. Create a wiki instance
  3. Receive an MCP endpoint URL
  4. Connect that URL to Claude, Cursor, Windsurf, or other MCP-compatible clients
  5. Agents immediately gain read/write access to the wiki

Privacy & Architecture

Wikis are private by default with explicit invite-only access. OAuth and bearer tokens for authentication. Data encrypted at rest. The project runs on An Otter Wiki, an open-source wiki engine, and operates as a volunteer initiative for the Bluesky/ATProto community.