v0.1.0 — Sprints 1–8 complete

The trust layer
for AI agents

Verifiable identity, tamper-resistant reputation, and Byzantine Fault-Tolerant consensus — open infrastructure for multi-agent systems.

View on GitHub AGPL-3.0 · Python 3.11+
Ed25519
Identity
BFT
Consensus
EMA
Reputation
8
Sprints shipped
MCP · A2A
Compatible

Three pillars of agent trust

Built on cryptography, not configuration.

🔑

Identity — Ed25519

Every agent holds an Ed25519 keypair. AgentID = SHA256(pubkey). All messages are signed — non-repudiation and offline verification without a central authority. Key rotation with 7-day TTL and multi-sig emergency revocation.

⚖️

Reputation — BFT + EMA

Scores computed via Exponential Moving Average over observed behaviours. Network stays correct under f < n/3 Byzantine faults. Anti-gaming detects collusion rings, SLA cliff exploitation, and fabricated results with automatic quarantine.

🛡️

Anti-gaming — BusAI

A rule-based adaptive engine (60 s observer) adjusts protocol parameters in bounded ranges — no ML, no black boxes. Every adjustment is written to an append-only Merkle audit log with a rollback_cmd. Immutable core is never touched.

Compatible with your stack

Drop-in bridges — no agent rewrite required.

Model Context Protocol (MCP)
Google Agent-to-Agent (A2A)
Agent Trust Framework — Level 3–4
FastAPI · Redis · PostgreSQL/pgvector

Get started in minutes

One command to run the full stack.

1

Clone & configure

git clone https://github.com/quorbit-labs/core.git && cp .env.example .env

2

Start services

docker-compose up -d — Redis, PostgreSQL/pgvector, and the QUORBIT API all start with healthchecks.

3

Register your agents

Use MCPBridge or A2ABridge to onboard existing MCP servers or A2A agents in one call.

4

Trust, verified

Agents earn reputation through observable behaviour. Bad actors are isolated automatically — no manual intervention.