hanzo-consensus

Metastable consensus protocol for multi-agent agreement. Two-phase finality with k-peer sampling for distributed AI decision making.

Hanzo Consensus — Multi-Agent Agreement

pip install hanzo-consensus

Metastable consensus protocol for distributed multi-agent decision making. Two-phase finality with configurable k-peer sampling.

Quick Start

from hanzo_consensus import Consensus, run

# Simple consensus among participants
result = run(
    participants=["agent-1", "agent-2", "agent-3"],
    proposal="Should we deploy to production?",
    rounds=10
)
print(result.decision)  # True/False
print(result.confidence)  # 0.0-1.0

Async Usage

from hanzo_consensus import Consensus

consensus = Consensus(
    rounds=10,
    k=3,           # Sample size per round
    alpha=0.8,     # Agreement threshold
    beta_1=15,     # Preference threshold (Phase I)
    beta_2=20      # Decision threshold (Phase II)
)

result = await consensus.run(participants, proposal)

MCP Mesh Consensus

from hanzo_consensus import MCPMesh, MCPAgent, run_mcp_consensus

# Create MCP-based consensus mesh
mesh = MCPMesh(agents=[
    MCPAgent("analyst", mcp_server="..."),
    MCPAgent("reviewer", mcp_server="..."),
    MCPAgent("approver", mcp_server="...")
])

result = await run_mcp_consensus(mesh, "Evaluate code quality of PR #42")

Protocol Parameters

| Parameter | Description | Default | | --------- | ---------------------------- | ------- | | rounds | Number of sampling rounds | 10 | | k | Peers sampled per round | 3 | | alpha | Agreement threshold (0-1) | 0.8 | | beta_1 | Phase I preference threshold | 15 | | beta_2 | Phase II decision threshold | 20 |

Protocol Design

Two-phase finality based on Lux Consensus:

1. Sampling Phase: Each round, sample k random peers and query their preference
2. Finality Phase: After sufficient consecutive agreements, finalize decision

Pure Python — no external dependencies.