Skip to content
nexus-agents

Governance substrate for AI coding agents

The agents do the engineering. nexus-agents holds them to the rules.

It reviews their work with adversarial consensus panels, keeps the rules they follow drift-checked, and writes everything they touch to a tamper-evident audit chain. It runs as an MCP server inside Claude Code, Codex, Gemini CLI or OpenCode.

npm install -g nexus-agents
$ nexus-agents vote --proposal "Add a 'nexus-agents tour' interactive walkthrough command that runs after 'setup'..."

Collecting votes from 7 agents (timeout: 300s each)...

✓ Software Architect:   APPROVE (82%)
✓ Security Engineer:    APPROVE (90%)
✓ Developer Experience: APPROVE (88%)
✗ AI/ML Engineer:       REJECT  (75%)
✓ Product Manager:      APPROVE (82%)
✗ Contrarian Analyst:   REJECT  (95%)
✗ Scope Steward:        ERROR   — MCP error -32001: Request timed out

Approval: 66.7%
Threshold: simple_majority
Result: APPROVED
Fig. 01 — Abridged from a real seven-voter run on 2026-05-17: the banner, the vote counts and the timing are cut, the rest is verbatim. One seat timed out and is reported as errored, not counted as a vote. Full transcript below.

Four ways into the docs

What it enforces

Adversarial review
A seven-seat consensus_vote panel, including a contrarian seat, decides real forks. Thresholds are counted over the voters who answered, and a change to the governance substrate needs a whole panel. There are 5 aggregation strategies, and every decision is recorded with each voter's reasoning.
Rules that cannot drift silently
The governance blocks in the rule files every agent reads are generated from the code, and CI fails when a hand edit diverges from the source.
A tamper-evident audit chain
Every decision is appended to a hash chain you can check with verify_audit_chain. Tamper-evident, not tamper-proof: the threat model says exactly what that buys.
Closed-loop telemetry
Outcomes feed back into model routing, within bounds, so a 12-stage router learns which of the 5 CLI adapters handles which task.

In the box: 47 MCP tools, 12 expert roles, 34 skills, 8 execution strategies, 8 memory backends, and a research registry of 176 papers from 9 discovery sources.

A real run, in full

$ nexus-agents vote --proposal "Add a 'nexus-agents tour' interactive
    walkthrough command that runs after 'setup'..."

Nexus Agents Consensus Vote
============================

Proposal: Add a 'nexus-agents tour' interactive walkthrough command that runs after 'setup'...

Collecting votes from 7 agents (timeout: 300s each)...

Votes

  ✓ Software Architect:   APPROVE (82%)
  ✓ Security Engineer:    APPROVE (90%)
  ✓ Developer Experience: APPROVE (88%)
  ✗ AI/ML Engineer:       REJECT  (75%)
  ✓ Product Manager:      APPROVE (82%)
  ✗ Contrarian Analyst:   REJECT  (95%)
  ✗ Scope Steward:        ERROR   — MCP error -32001: Request timed out

Summary

  Approve:  4
  Reject:   2
  Abstain:  0
  Errored:  1
  Approval: 66.7%
  Threshold: simple_majority

Result: APPROVED

Completed in 188487ms
Fig. 02 — Verbatim output of a real seven-voter run on 2026-05-17, a proposal from the website backlog. One seat timed out and is reported as errored, not counted as a vote.