C3PO — The Protocol Institute

C3PO is the Protocol Institute's conversational research assistant — a RAG-based AI agent trained on the Institute's full research corpus. It lets researchers, practitioners, and curious readers query, synthesize, and explore the protocol research library through natural language, with citations. Try it →

The corpus spans over 12,000 indexed passages from academic papers, working papers, essays, talks, datasets, games, interviews, fiction, Discord discussions, SIG meeting records, and Protocolized magazine content — the majority originating from the Summer of Protocols program (2023–2024), and growing continuously. C3PO makes that corpus conversationally accessible: ask it a question about protocol theory, a specific paper, a design problem, or a cross-domain connection, and it will draw on the corpus to respond — with citations.

What C3PO does

C3PO is a RAG (retrieval-augmented generation) system. When you ask a question, it searches the embedded research corpus for the most relevant passages, then generates a response grounded in those sources. It always cites what it found. It does not answer general knowledge questions outside the corpus, and it is transparent about the limits of what it knows.

Intended uses include:

Finding relevant research on a specific protocol domain (medical, diplomatic, software, financial, urban)
Synthesizing what the corpus says about a theoretical question (hardness, formalization, enforcement, capture)
Discovering connections across papers and disciplines that are hard to find by browsing alone
Getting oriented in the literature before diving into primary sources

Status

C3PO is live in public beta at c3po.vgr-702.workers.dev. The index currently spans 12,000+ vectors across five namespaces: the Protocolized PDF library and Substack archive, Protocol Institute Discord discussions, SIG (Special Interest Group) meeting records, and video transcripts. The corpus grows as new content is published and ingested.

C3PO is also available as an MCP (Model Context Protocol) server, letting Claude Code and Claude Desktop users search the corpus and ask questions directly from their AI coding environment — without opening a browser. See the how it works page for connection instructions.

Technical approach

C3PO uses Voyage AI embeddings (voyage-3, 1024 dimensions), Pinecone as the vector store, and Claude Sonnet as the language model. It runs as a Cloudflare Worker serving a web UI, a rate-limited REST API, and an MCP server endpoint from a single deployment. The source code is at github.com/vgururao/c3po.