Multi-agent code review · Now in session

Six minds.
One verdict.

AI Council is a multi-agent code review engine that reviews your code as you write it. Six specialized agents debate every change, vote with confidence scores, and hand down one reasoned ruling, so what you ship is the best it can be. Pair it with any model and the Council makes it smarter.

Retain the Council Read the documentation

The Verdict

Watch a ruling come down.

Proceedings docket № 2026-0812
$ ai-council review --diff --branch=main --pretty
  1. Senior Developer Approve 0.92

    Code follows best practices, well-structured.

  2. Security Engineer Approve 0.88

    No security vulnerabilities detected.

  3. Performance Engineer Revise 0.75

    Consider memoizing the expensive computation.

    const result = useMemo(() => compute(data), [data]);
  4. Architect Approve 0.85

    Good separation of concerns.

  5. Gemini Approve 0.90

    Logic is sound, edge cases handled.

Final ruling Approved 4 approve · 1 revise · confidence 0.86

Capabilities

The full instrument.

01

Multi-Agent Consensus

6 specialized AI agents with unique perspectives collaborate to review your code, ensuring comprehensive coverage.

02

Automated Debates

When agents disagree, they engage in structured debates to defend their positions before the final ruling.

03

Confidence Scoring

Each vote includes confidence levels. Low-confidence agents automatically abstain, ensuring meaningful votes.

04

Specialized Reviews

Focus on security, performance, architecture, or general sanity with specialized agent configurations.

05

CI/CD Integration

Use in CI mode with exit codes for automated pipelines. Block merges on low confidence or rejections.

06

Fix Suggestions

When agents vote REVISE or REJECT, they can provide concrete code fix examples with syntax-highlighted, prettified output.

07

AST + Embedding Context

Opt-in pre-review pipeline extracts code units via tree-sitter AST and ranks them with CodeBERT embeddings for focused agent context.

08

Graph Context

Auto-discovers Neo4j, Memgraph, or HTTP graphs and injects neighborhood context around changed files, fail-open so reviews never break.

09

Deep Analysis

File-by-file review with structured findings: line-specific issues, severity levels, fix suggestions, and automatic project-rule injection.

10

Agent Review

Two-pass Claude Agent SDK pipeline: overview scan then deep per-file review with full codebase access via Read, Grep, and Glob.

11

MCP Server

Native integration with Claude and Cursor via MCP protocol. Review code directly from your favorite AI tools.

The Bench

Six members preside.

Six members, six distinct perspectives on every line of code you bring before them.

The Stone Owl

Senior Developer

Code Quality Expert

15+ years experience. Focuses on maintainability, best practices, and long-term sustainability.

Generalist

The Obsidian Sentinel

Security Engineer

Security Specialist

Identifies vulnerabilities, threat modeling, authentication issues, and secure coding practices.

Specialist

The Clockwork Head

Performance Engineer

Performance Specialist

Analyzes algorithmic complexity, resource usage, bottlenecks, and optimization opportunities.

Specialist

The Temple

Software Architect

Architecture Specialist

Evaluates design patterns, scalability, separation of concerns, and system boundaries.

Specialist

The Twin Crystals

Gemini Thinker

Alternative Reasoning

Powered by Google Gemini. Challenges assumptions and provides unique system-level perspectives.

Challenger

The Boulder

Pragmatist

Practical Problem Solver

Balances perfectionism with shipping. Sometimes "good enough" is the right answer.

Generalist

Procedure

Due process for every diff.

  1. Submit Code

    Run ai-council with your git diff or code snippet. The council receives your changes for review.

  2. Agents Analyze

    Each agent reviews from their unique perspective. Specialists check domain relevance before voting.

  3. Cast Votes

    Agents vote APPROVE, REVISE, or REJECT with confidence scores. Low confidence triggers abstention.

  4. Debate if Needed

    If Gemini dissents from majority, agents engage in structured debate defending their positions.

  5. Final Ruling

    The Judge synthesizes all votes and debates into a final recommendation with rationale.

Commission

Retain the Council.

One license. Six agents. Instant key. You bring your own model keys; your diffs never touch our servers. Checkout is handled securely by LemonSqueezy.

  • All 6 agents included
  • CLI, CI, and MCP for Claude and Cursor
  • License key delivered instantly
  • Manage seats and billing anytime

Already retained? Use My Account to update billing, seats, or retrieve your key.

License

Per seat, billed monthly

Standard

The full six-agent bench, including deepReview.

$4.99

Premium

Everything in Standard, plus agent mode.

$5.99

Buy License

Pick Standard or Premium at checkout. Your key usually arrives within a minute.

Inquiries

Questions for the record.

What exactly is AI Council?

A multi-agent code review engine. Six specialized AI agents, covering code quality, security, performance, architecture, and more, review your changes as you write them, debate disagreements, vote with confidence scores, and hand down one reasoned ruling. It runs from the CLI, in CI, or inside Claude and Cursor via MCP.

Does it work with the model and editor I already use?

Yes. The Council sits alongside any model you code with and supplements it with structured, multi-perspective judgment. It integrates with Claude and Cursor through the MCP server, works with any editor through the CLI, and blocks merges in any CI pipeline via exit codes.

Do I need my own API keys?

Yes. You bring your own Anthropic, OpenAI, or Gemini keys, and your license key unlocks the Council itself. At least one of OpenAI or Gemini is required; see the installation guide for setup.

Is my code stored anywhere?

No. Reviews run on your machine, and your diffs go directly from your machine to the model providers you configure with your own keys. We never receive, see, or store your code. See the privacy policy for details.

How do I get help, and how fast will you respond?

Email support@ai-council.tech and we'll respond within one business day. Licensing and billing questions are handled through My Account.

The Council is ready to convene.

Install via npm, add your license key, and bring every diff before the bench.

Purchase License 
# Install globally
npm install -g @mugzie/ai-council

# Set your license key
export AI_COUNCIL_LICENSE_KEY="your-key"

# Start reviewing!
ai-council review --diff --pretty