What is a Call Wiki?
A call wiki is a comprehensive hypertext wiki website created from video call transcripts and chat logs, assembled primarily by an AI system (for instance, Claude Code) under human direction. It transforms linear conversation recordings into a rich, navigable knowledge base.
Key Characteristics:
Content Structure:
- A multi-page hypertext website (say, ~90 pages for a 90-minute call) recounting topics, categories, quotes, and participants from video calls
- Organized using Markdown files with Double Square Bracket Links for wiki-style navigation
- Includes participant pages, topic pages, concept indexes, and background information on referenced subjects
- Contains hubs like "Participants Hub" and "Concept Index" for easy navigation
Creation Process:
- Built by Claude Code (an agentic AI system) that semi-autonomously works with transcript and chat documents
- The AI has significant autonomy in deciding structure and content organization, guided by high-level human instructions
- Key prompts focus on: analysis, synthesis, comprehensive summary, organization of people/books/organizations, and creating "orphan pages" (linked concepts that get fleshed out with explanations)
Purpose & Philosophy:
- Related to Pete's "Field of Sheaves" project - exploring how humans and AI can co-create structured, evolving knowledge bodies
- Transforms ephemeral conversations into persistent, explorable knowledge artifacts
- More experimental and exploratory than traditional meeting notes or summaries
- Emphasizes hypertext connections between ideas, making conversation topics discoverable and interconnected
Distribution:
- Typically hosted as a website published with MarkPub; backed by GitHub repositories with downloadable ZIP files
- Typically include disclaimers like "WORK IN PROGRESS: PAGES NOT YET REVIEWED BY HUMAN EXPERTS. VERIFY CLAIMS AND CONSULT ORIGINAL SOURCES"
- Meant to be finite artifacts of works-in-progress more than finished publication
Call wikis so far have been an instance of Pete's exploration of AI-assisted knowledge management - they're fun, somewhat useful, experimental checkpoints rather than finished products.