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
OpenClaw Agent Gains Phone Call Capability Through Custom Skill
A developer created a custom skill for self-hosted OpenClaw agents that enables phone call functionality, allowing the agent to initiate calls based on triggers like build completions or server outages. The implementation provides voice interaction with full chat capabilities including web searches and alert setup.
E2a: Open-Source Email Gateway for AI Agents with SPF/DKIM Verification and Webhook/WebSocket Delivery
E2a is an authenticated email gateway for AI agents that verifies SPF/DKIM on inbound mail, delivers via webhook or WebSocket, and supports outbound email with human-in-the-loop approval.
Testreel: Programmatic Demo Video Generation with Claude Code
Testreel is an npm package that generates polished product demo videos from JSON, YAML, or Playwright interaction descriptions. It creates webm/mp4/gif videos with cursor overlays, click ripples, and gradient backgrounds.
Bridge Claude Code to Chat Apps for Remote Interaction
A GitHub project called cc-connect bridges Claude Code to messaging platforms like Slack and Telegram, allowing remote interaction without exposing your local machine. The agent runs locally while a small bridge relays messages between the agent and chat apps.