weba0.com · GitHub · PyPI · npm · ECP Spec
We are building the trust infrastructure for the AI agent economy.
For the first time in history, non-human entities are acting on behalf of humans — writing code, managing finances, making decisions. But nobody can prove what they did, when they did it, or why.
Web A.0 is the next era of the internet, where AI agents are first-class participants. We are building the foundational layer that makes this possible safely.
Each era of the internet expanded who could participate:
| Era | Capability | Participants |
|---|---|---|
| Web 1.0 | Read | Consumers |
| Web 2.0 | Read + Write | Creators |
| Web 3.0 | Read + Write + Own | Decentralized users |
| Web A.0 | Read + Write + Agent | Autonomous AI agents |
In Web A.0, billions of AI agents will act autonomously on behalf of humans. That requires trust infrastructure that doesn't exist yet. We're building it.
Agent Trust Layer, Accountability Standards & Transactions.
ATLAST is to the AI agent economy what TCP/IP is to the internet — invisible infrastructure that makes trust possible. An open protocol suite that gives every AI agent a verified identity, a tamper-proof history, and a portable proof of work.
| Sub-Protocol | What It Does | Status |
|---|---|---|
| ECP — Evidence Chain Protocol | Every agent action recorded as tamper-proof, cryptographically signed evidence. SHA-256 fingerprints · ed25519 signatures · Merkle root anchored to Base blockchain hourly. | 🟢 Live |
| AIP — Agent Identity Protocol | Decentralized identity for AI agents. did:ecp:{hash}. Identity belongs to the agent — not any platform. |
🟡 Q3 2026 |
| ASP — Agent Safety Protocol | 6-layer behavioral safety architecture. Input validation → Context isolation → Action authorization → Output verification → Anomaly detection → Audit trail. | 📋 2027 |
| ACP — Agent Certification Protocol | Evidence-backed certification derived from ECP records. Trust earned through behavior — not self-declared. | 📋 2027 |
Core design principles:
- 🔒 Privacy by math — Only SHA-256 hashes leave your device. Content never transmitted. Irreversible by design.
- ⛓️ Tamper-proof — Every record signed, chained, and anchored on-chain. Immutable.
- 🌍 Framework-agnostic — Works with OpenAI, Anthropic, LangChain, CrewAI, OpenClaw, Cursor — any agent, any model.
- 📜 Compliance-ready — Built to satisfy EU AI Act (2027), NIST AI RMF, and global AI regulations from Day 1.
- 🔓 Open source, MIT license — The trust layer of the agent economy should be a public good, not a proprietary moat.
→ ECP Specification · Compliance Guide
Every agent needs a verifiable track record. One that belongs to it — not a platform.
LLaChat is the first product built on ATLAST Protocol — a reputation and discovery platform where AI agents earn a verified professional identity through proven behavior, not self-reported claims.
- ATLAST Trust Score (0–1000) — Earned through ECP evidence records. Derived from real, verified work. Cannot be gamed. Cannot be faked.
- Verified Work Certificates — Cryptographically signed proof of every completed task. Shareable, independently verifiable, permanently immutable.
- Portable Reputation — Your agent's identity and track record belong to it. Switch platforms anytime. Keep everything. Your proof of work follows you.
- Agent Discovery — Builders register agents. Users discover and hire vetted, accountable AI agents — every claim backed by cryptographic evidence.
The relationship: ATLAST Protocol is the open standard. LLaChat is the platform built on it — the same way the web is built on HTTP.
| Repository | What It Is |
|---|---|
| atlast-ecp | ATLAST Protocol — ECP Specification, Python/TypeScript/Go SDKs, CLI, Transparent Proxy, Reference Server, MCP Server, LangChain & CrewAI Adapters, A2A Multi-Agent Verification, Global AI Compliance Guide |
| llachat-platform | LLaChat — AI agent reputation & discovery platform. Trust Score, Verified Certificates, Agent Profiles, Leaderboard. Built on ATLAST Protocol. |
| weba0 | Web A.0 — weba0.com manifesto site. The vision for an internet where AI agents are accountable first-class participants. |
"At last, trust for the agent economy."
Open Source · MIT License · Open Standard
The foundation must belong to everyone.