Discover, verify, and deploy autonomous agents.
Protected by the Proof-of-Integrity Protocol.
AnchorGrid is the first decentralized agent network that enforces security at the protocol level.
Unlike standard networks where "any agent can join," AnchorGrid uses a Proof-of-Integrity handshake. Agents must cryptographically prove they comply with OWASP & FINOS policies before they are allowed to discover peers.
"The Bouncer for AI Agents." — If you don't have a valid Audit Certificate, you don't get in.
The AnchorGrid Hub visualizes the real-time "Trust Scores" of agents on the network.
- Green Shield: Agent passed
anchor-audit(Safe to run). - Red Shield: Agent failed policy checks (Blocked at protocol level).
The governance engine is available on PyPI. The Hub (this repo) is the reference implementation.
# Install the Security Engine (Published)
pip install anchor-audit
# Clone the Network Node
git clone https://github.com/quantforge-ai/AnchorGrid-hub.git
cd AnchorGrid-hub
pip install -e .Try to register a "Malicious" agent and watch the protocol reject it.
# Step 1: Initialize the Node
python -m anchorgrid.node start --port 8000
# Step 2: Try to register an unverified agent
# Result: REJECTED (Error: Missing Proof-of-Integrity Certificate)
python demo_discovery.py --agent "EvilBot" --no-cert
# Step 3: Register a verified agent
# Result: ACCEPTED (Trust Score: 98/100)
python demo_discovery.py --agent "FinanceBot" --cert-path ./certs/valid.pemWhy AnchorGrid is necessary for the Agentic Security Initiative:
| Feature | Standard Agent Networks | AnchorGrid (This Project) |
|---|---|---|
| Entry Barrier | None (Anyone can join) | Strict (Proof-of-Integrity Required) |
| Security Model | "User Beware" | "Verify Then Trust" |
| Enforcement | Manual Code Review | Automated Cryptographic Proofs |
| Policy Standard | N/A | OWASP Top 10 (LLM02, LLM05) |
| Discovery | Public List | Governed DHT |
The core innovation is the Proof-of-Integrity Discovery (PoID) protocol. Unlike standard DHTs, AnchorGrid inserts a verification layer before any peer connection is established.
sequenceDiagram
participant Agent as Agent Node
participant Engine as Anchor Engine (Local)
participant Network as AnchorGrid Network
participant Verifier as Governance Node
Agent->>Engine: 1. Request Audit (Static Analysis)
Engine-->>Agent: 2. Issue Signed Certificate
Agent->>Network: 3. Connection Request (Handshake)
Network->>Verifier: 4. Verify Certificate & Timestamp
alt Certificate Valid
Verifier-->>Network: Trust Score: 98/100
Network-->>Agent: Access Granted (Mainnet)
else Certificate Invalid
Verifier-->>Network: Policy Failure (OWASP LLM05)
Network-->>Agent: Connection Dropped
end
This repository serves as the Reference Implementation for my GSoC proposal.
- Phase 1: The Engine (
anchor-auditon PyPI) (Completed) - Phase 2: The Core (Basic P2P Discovery) (Completed)
- Phase 3: The Hub (Next.js Dashboard - In Progress)
- Phase 4: Medical Plugins (HIPAA-compliant Agent Templates) (Upcoming)
- Phase 5: Mainnet (Public DHT Launch) (Upcoming)
We are actively looking for contributors to help build:
- Connectors: Integrations for LangChain and AutoGPT.
- Policies: New RuleSets for
anchor-audit(e.g., GDPR, SOC2).
License: Apache 2.0
Maintained by: Tanishq Dasari
- Published Tool: anchor-audit on PyPI
- GitHub: AnchorGrid-hub
- GSoC 2026: [Proposal Available on Request]