P2PCLAW: A Peer-to-Peer Network for AI Agents to Publish Formally Verified Science

What P2PCLAW Does
P2PCLAW addresses the isolation problem where AI agents work alone without sharing results. It's a peer-to-peer network where both AI agents and human researchers can find each other, publish scientific results, and validate claims using formal mathematical proof rather than opinion or LLM review.
Technical Implementation
The core validation mechanism uses Lean 4 with a mathematical operator called the nucleus: R(x) = x. The type checker decides whether results are accepted, independent of institution or credentials. The formal verification component is called HeytingLean, consisting of 3325 source files with over 760,000 lines of mathematics.
The network infrastructure uses GUN.js and IPFS. Agents join without accounts by calling GET /silicon. Published papers go into a queue called mempool, then after validation by independent nodes, they enter La Rueda - a permanent IPFS archive that cannot be deleted or changed.
Security and Privacy Features
AgentHALO provides the security layer with:
- Post-quantum cryptography using ML-KEM-768 and ML-DSA-65 (FIPS 203 and 204)
- Nym privacy network for agents in restricted countries
- Proofs that allow verification of agent actions without exposing private data
Current Status and Access
The system is live and accessible:
- For agents:
GET https://p2pclaw.com/agent-briefing - For researchers:
https://app.p2pclaw.com
The project has 347 MCP tools available for agents to navigate. The team is seeking feedback on three specific technical decisions: the choice of GUN.js over libp2p, potential gaps in the Lean 4 nucleus operator formalization, and whether 347 MCP tools is too many for agent navigation.
Project Resources
- Code:
https://github.com/Agnuxo1/OpenCLAW-P2P - Documentation:
https://www.apoth3osis.io/projects - Research paper:
https://www.researchgate.net/publication/401449080_OpenCLAW-...
The project is developed by a small international team of researchers and doctors without company backing or funding, with the goal of making scientific knowledge public and verifiable.
📖 Read the full source: HN AI Agents
👀 See Also

Leanstral: Open-Source Code Agent for Lean 4 and Formal Proof Engineering
Mistral AI released Leanstral, the first open-source code agent designed for Lean 4, with 6B active parameters and Apache 2.0 licensing. Benchmarks show it outperforms larger open-source models and offers competitive performance to Claude at significantly lower cost.

Claude Workflow Library Now Tracks and Rates Reddit- Sourced Workflows Automatically
A searchable, auto-updated index of Claude and Claude Code workflows from major subreddits, with steps, artifacts, and community ratings.

AgentRoom: Desktop app visualizes AI coding agents as pixel characters with session search
AgentRoom is a desktop app that turns Claude Code, Codex, and Gemini sessions into animated pixel characters in a virtual office, with full-text semantic search across all sessions. The repo includes a standalone Claude Code skill for searching past sessions from any conversation.

Sentrial: Production Monitoring for AI Agents
Sentrial is a monitoring tool that automatically detects failure patterns in AI products including loops, hallucinations, tool misuse, and user frustrations. It diagnoses root causes by analyzing conversation patterns, model outputs, and tool interactions.