Open source · Free forever · Apache 2.0

Stop reviewing what
a machine can catch.

Manta puts 20 review agents in your git workflow. Secrets, injection flaws and unsafe migrations are blocked before they land in your repo.

Three agents read every commit. The rest run at push, routed by what the change actually touches. Works with Claude Code, OpenAI Codex, Gemini CLI and GitHub Copilot — the hooks detect whichever you have installed.

payments-api — git commit
$ git commit -m "feat: add referral payouts"

  Manta Pre-Commit Review
  Reviewing 5 staged file(s)…
  Running security-sentinel, code-quality,
  db-migration-guardian

  CRITICAL  src/routes/payout.ts:42
            Unparameterised SQL from req.query.userId · CWE-89

  CRITICAL  migrations/0042_payouts.sql:7
            NOT NULL on a 4.2M-row table, no rollback

  WARNING   src/routes/payout.ts:19
            Validation duplicated from refund.ts:31

  ✗ COMMIT BLOCKED · 2 critical findings
  Run /fix for AI-generated fix suggestions.

What runs, and when

Reviewed where it is cheapest to fix

The split is not arbitrary. Three agents run at commit — the checks where committing is itself the harm, because a secret in the history stays in the history. Everything else waits for push, where catching something late costs a rebase rather than a rotated credential.

git commit · 3 agents

security-sentinel — secrets, API keys, SQL injection, XSS, auth bypass, OWASP Top 10.
code-quality — DRY, complexity, dead code, naming, missing error handling.
db-migration-guardian — table locks, missing rollbacks, unsafe NOT NULL — only when a migration is staged.

git push · broader scope

The same checks across the whole branch, plus perf-analyzer — N+1 queries, memory leaks, blocking operations, bundle bloat. A CRITICAL blocks the commit; a WARNING blocks the push.

On demand · 21 commands

/audit, /security-scan, /write, /scaffold, /ui, /blueprint, /wiki, /explain, /debt, /generate-tests — and the RPI workflow below.

A review takes minutes, not seconds. The agents read your code rather than lint it, and they bill to your own AI subscription — MANTA_MODEL and --depth are the two dials that decide how much you spend on each pass.

What's included

20 agents. 21 commands. Nothing to configure.

Security, code generation, quality enforcement and documentation, in one install. Any language — TypeScript, Python, Go, Rust, Ruby, Java, Kotlin, C/C#/C++, PHP, Swift.

Security on every commit

OWASP Top 10, hardcoded secrets, API keys, SQL injection, XSS and auth bypass — caught and blocked before they reach your repo.

Complete implementations

/write "feature" produces production-ready code — rate limiting, auth wiring, validation, pagination and transactions included. No TODOs, no stubs.

Quality & performance

DRY violations, cyclomatic complexity, N+1 queries, memory leaks and unsafe migrations, caught before they compound into something harder to fix.

Docs that stay current

/update-docs keeps README and CHANGELOG in step with your commits; /blueprint maps an unfamiliar codebase in one pass.

Before you write a line

Research → Plan → Implement, with a GO/NO-GO gate

For anything non-trivial, the cheapest bug to fix is the one you decide not to build. Six agents evaluate the feature first, and the gate will tell you not to proceed.

1 · Describe

Write rpi/{slug}/REQUEST.md in plain language.

2 · Research

/rpi-research — six agents, ending in GO / CONDITIONAL GO / DEFER / NO-GO.

3 · Plan

/rpi-plan — product requirements, UX, engineering spec, phased roadmap.

4 · Implement

/rpi-implement — phase by phase, with a validation gate after each.

Free edition and Enterprise

Everything here is free. This is what the paid edition adds.

Manta stays free and open source — that is not a trial. Enterprise adds the governance layer: the agents a regulated team needs, and a hosted record of what the gate decided.

 MantaManta Enterprise
Agents2033
Commands2132
Review at git commit3 agents3 agents
Review at git push3–4 agents, trigger-routedup to 9 agents, trigger-routed
A warning at pushblocksblocks only when two agents agree
Code generation — /write, /scaffold, /ui
Research → Plan → Implement, with a GO/NO-GO gate
Security scanOWASP + secrets+ dependency CVEs, licence conflicts, dead deps
Spec governance — spec-guardian
Compliance — GDPR, HIPAA, PCI-DSS, SOC 2
Zero-trust — IAM, mTLS, RBAC, token TTLs
Observability checks on new endpoints
Test-coverage enforcement
Health scoring and trends across audits
Penetration testing — /penteston demand
Log analysis — /log-analysis
Hosted enforcement record and evidence bundles
LicenceApache 2.0, free foreverCommercial subscription

Enterprise adds 13 agents; 11 of them are served from the control plane while a subscription is active. Counts are generated from the shipped rosters, not typed by hand.

Compliance enforcement, zero-trust checks, health scoring, penetration testing, and an audit-ready record of every block and every bypass.

Explore Enterprise →

Get started

One command. Any project.

Install into an existing repo or start fresh. The next git commit triggers the pipeline — no account, no service to run, nothing sent anywhere.

$ gh repo clone mantacron/manta /tmp/manta && bash /tmp/manta/scripts/install.sh && rm -rf /tmp/manta

Then open your AI tool and run /init for a new project, or /audit to get a health report on an existing one. Codex and Gemini take the same prompts directly: codex "$(cat .claude/commands/audit.md)".

Questions or feedback? ray@mantacron.com