Home

Decentralized Runtime Integrity Engine

Proof your remote agent hasn't been touched. No trust in the remote machine.

A verifiable integrity framework for remote agents: on-chain anchored tamper detection with golden-baseline verification. Remote agents (AI bots, edge workers, enterprise daemons) run where you can't see them: an attacker with file access can patch the binary, edit config.json, or flip an in-memory limit. This engine does not prevent the edit, it proves it within one 5s interval (pipeline average ~4ms, worst 24ms, measured over 20 trials) and flags STALE if no fresh anchor arrives after 12s.

Chain proves timeline. Baseline proves what good looks like. You need both: chain alone would accept an honest Checker that anchors a hash of the attacker's edited file; baseline alone cannot prove order. The Verifier demands H_recomputed == H_chain == H_baseline.

Quick start

git clone https://github.com/rajit2004/Decentralized-Runtime-Integrity-Engine-for-Remote-Agents.git
cd Decentralized-Runtime-Integrity-Engine-for-Remote-Agents
javac -d out (Get-ChildItem -Recurse java/src/*.java)
java -cp out integrity.Main

Open http://localhost:8080 - you should see GREEN flowing with cycle + tx. A fresh clone boots GREEN immediately: the golden baseline is committed and keys/ auto-generate on first run.

Wiki pages

Page	What it covers
Architecture	Components, data flow, mermaid diagram
How It Works	Enrollment + heartbeat phases, the six checks, state machine
Verdict Codes	Every GREEN/RED reason, attack table, contract API
Demo Guide	Running each live demo end to end
Testing & CI	86-check SelfTest and the CI gates
Open Limits	What the system does and does not cover

Tech in one line

Pure JDK Java 17+ (no Maven/Gradle), Ed25519 + SHA-256, Solidity Integrity.sol on a Hardhat local node, JDK HttpServer dashboard. 21 Java files, one job each.