CLI · API · MCP · Wrapper · Agent SDK

One engine.
Every interface.

consensus.tools works however your pipeline works. Use guards as a library, run the CLI in CI/CD, call the REST API from any language, connect via MCP for Claude and Cursor, or wrap functions directly with the runtime firewall.

Core Concept

What is a Decision Firewall?

A Decision Firewall sits between “proposal” and “execution.”

Decision Flow
Instead of:
Agent→Execute
You get:
Agent→Propose→Guard→Validate→Quorum→Finalize
No silent drift.
No unbounded delegation.
No single-agent authority escalation.

Every decision can require:

Weighted quorums
Human-in-the-loop approvals
Multi-model validation
Reputation-based voting
Block or flag low-quality behavior
Custom programmable policy logic

This applies to:

GitHub Pull Requests
CI/CD deploy approvals
Multi-model inference pipelines
Autonomous trading systems
Warehouse robotics coordination
Production workflow automation
Agent-to-agent governance
Use Case

Applying to Pull Requests

Example: High-Risk Code Change

PR Consensus Flow
Policy
personas_required: 3
human_reviewers: 1
weighted_quorum: 60%
block_on_fail: true
flag_for_review: test_coverage | policy_violation
Flow
1
Agent submits PR
2
PR enters consensus board
3
Personas review and vote
4
Human reviewer confirms
5
Quorum reached
6
Merge unlocks automatically
Outcome: Deployments become programmable governance layers.
Use Case

Applying to Multi-Model Systems

Example: Multi-Model Decision Arbitration

Multi-Model Arbitration
Actors
GPT-4 class reasoning model
Lightweight local model
Retrieval system
Risk classifier
Human operator (optional)
Policy
model_agreement: 2-of-3
risk_model_veto: true
human_override: high_risk
scoring: reputation_weighted_confidence
Flow
1
Proposal generated
2
Models independently score
3
Votes aggregated
4
Policy threshold check
5
Finalized decision emitted with trace_id
Result: Collective machine intelligence. No single model decides alone.
Infrastructure

Consensus Policy Templates

Consensus is not a black box. Policies are TypeScript templates built with createPolicyTemplate(). Extend any of the 9 base algorithms, override thresholds, add pre-checks, and register custom resolvers.

Open Source

Composable & Open Source

Every policy is:

Open source
Forkable
Extensible
Programmable

You can:

Create custom quorum math
Modify reputation and weighting logic
Add block and flag conditions
Integrate OpenTelemetry tracing
Define identity requirements per board
Run locally or self-host

The engine is Apache-licensed.

Fork it. Compose it. Ship your own consensus policy layer.

Deployment

Privacy & Deployment Modes

Run Locally
Fully offline
No account required
UUID-based agent identities
Air-gapped capable
SQLite-backed audit ledger
Self-Hosted / CI Integration
Docker or bare-metal deployment
CI/CD pipeline integration
GitHub Actions native support
OpenTelemetry tracing
Custom guard configurations

Agents don't need hype.

They need guardrails.

Consensus is the programmable decision firewall.

Interfaces

Five ways to integrate

Use consensus-tools however your pipeline works. CLI for CI/CD, SDK for TypeScript apps, REST API for any language, MCP for LLM agents, or the wrapper for function-level gates.

CLI

Run guards and query the board from your terminal or CI/CD pipeline. Pipe JSON in, get decisions out.

View docs →
terminalbash
# Evaluate a code merge action through the guard
consensus-tools guard evaluate \
  --type code_merge \
  --payload '{"files":["src/auth/login.ts"],"tests_passed":true}' \
  --format json

# Post a consensus job for PR review
consensus-tools jobs post \
  --title "Review PR #142" \
  --policy WEIGHTED_REPUTATION \
  --json

# Query the audit trail for blocked actions
consensus-tools board audit \
  --filter "decision=BLOCK" \
  --since 24h \
  --format table
Open Source

Built in the open

The consensus.tools CLI, guard packages, and agent skill.md are production-grade and open source. The full guard pipeline runs locally with no external dependencies. Self-host or integrate into existing CI/CD.

What consensus.tools is

Core Utility

Guard layer

Pre-execution governance for high-risk autonomous actions.

Consensus engine

Policy-driven resolution of multi-agent decisions.

Audit trail

Every decision is written as an immutable board artifact.

Pipeline orchestration

Integrates into CI/CD, agentic workflows, and multi-model systems.

Non-Goals

Payment system

No credits, tokens, or economic mechanisms.

Token platform

No staking economy or token issuance.

Blockchain

Runs on a local SQLite ledger — no distributed consensus chain.

Withdrawal platform

No financial rails or fund management.

Ready to deploy
consensus?

Install the CLI locally and ship your first guard pipeline in minutes. The engine is open source and production-grade.

Open-source / Apache 2.0 Licensed / Production-grade