// independent.trust.layer

The independent referee
for AI-written code.

WardRail checks every change your agents make against your project's own contract: deterministic security scans plus an LLM verdict on whether it did what was asked and nothing it shouldn't. It runs in your CI, on your key. We never see your code.

0
// skills & loops
0
// languages graphed
£4.99
// pro / month
wardrail · live · src/session.ts

// what.is.wardrail

Agents ship code faster than any human can review it.

Wardrail is the safety cockpit around your coding agent. You agree a set of rules for the project, a contract, and Wardrail enforces it on every change: a deterministic security scan, an independent LLM verdict on whether the agent did what was asked and nothing it shouldn't, a live code graph your agent can query, and a running record of where the repo has drifted. It's the referee that holds the line between moving fast and shipping slop, and it does that without ever seeing your code.

security scanproject contractindependent verdictcode graphdrift trackingthe WardenBYOK vault238 templates

// paste.a.diff

Paste a diff. Get a verdict.

A real security scan, running in your browser with no network and no key. Findings and the trust score update as the code changes. The full product adds the LLM verdict against your committed contract on top.

editable · in-browserdemo
0drift
1 FINDING

in-browser security scan

⚠ L2 · string-built SQL — injection risk

const sql = "SELECT * FROM users WHERE id = " + id;

// code.graph

Query your repo, don't read it.

A native graph across TS/JS, Python, PHP and Go. It shows the blast radius of any change, everything that transitively depends on it. The Warden answers the same questions in plain English.

query ›who calls createSession?

preset queries

Reading the graph…

// drift.timeline

Drift never accumulates in the dark.

The repo line wanders from your contract as agents commit, and every verdict pulls it back onto the rail. When a secret is committed the scene spikes red, a signed record drops into the ledger, then it's corrected.

✓ two lines converge. Drift caught and corrected, every time.

trust ledger

No records yet.

// the.warden

Ask your project anything.

The Warden knows your contract, findings and code graph. Ask what changed, what's affected, or whether agents are allowed somewhere, and it answers from your repo with the artefacts attached. Turn on Propose changes and it can draft a guardrail for you to approve.

  • Query the code graph in plain English
  • Explain any verdict, finding or drift
  • Propose guardrails you sign off on, never enforced silently

Warden

runs in your browser with your key · proposes, you approve

Ask Warden…
Send

// with.vs.without

Same four PRs. Two very different weeks.

// without wardrail

// with wardrail

// the.console

One cockpit for the whole trust surface.

AST security findings

✓ no hardcoded secrets

✓ no injection surface

⚠ raw SQL — report.ts:41

scanned 214 files · 0.8s

// templates

A library of skills & loops your agents pull from.

Skills teach an agent to do one thing your way, like adding a REST endpoint, writing a safe database migration, or shipping an accessible component. Loops are recurring jobs, like a nightly security sweep or a dependency audit, that run and report on their own. Browse, add to your library, copy, or commit straight to your repo.

BackendFrontendSecurityTestingDevOpsDatabaseAI agentsAPI designObservabilityPerformanceCachingCloudDesign SystemsData engineering+22 more
skillBackend

Add a REST endpoint

add to library
skillDatabase

Write a safe database migration

add to library
loopSecurity

Nightly security sweep

add to library
skillBackend

Harden a payment webhook

add to library
skillFrontend

Ship an accessible component

add to library

// who.it.is.for

One question. Traced to an answer.

you › did my agent break anything in this PR?

verdict › 91 · pass

finding › nothing above low severity

rule › matches contract v3

commit › a1b2c3 · safe to merge

why £4.99 · Ship agent-written code without becoming the bottleneck reviewer.

// vs.the.alternatives

What raw agents and generic linters miss.

CapabilityWardRailRaw agentGeneric linter
Independent verdict on every change
Enforces your committed contract
file:line security scan
Queryable code graph
Drift history
BYOK · zero code egress~~
Signed, verifiable records

// bring.it.to.your.workflow

Four ways in. Pick where your agents live.

GitHub Action

An independent verdict on every PR, in your own CI with your own key.

setup guide

// your.code.your.machine

We built the trust tool to be trustworthy.

A tool that reviews your code shouldn't be the thing you have to worry about. So Wardrail is designed to know as little as possible, and to run where you already trust the environment.

  • Bring your own key, zero-knowledge

    Model keys are encrypted in your browser. We can't read them, and every inference runs on yours.

  • Runs in your CI, on your key

    The PR verdict executes inside your own pipeline. Our servers are never in the path of your code.

  • Read-only GitHub scope

    The app reads the repo you connect. It never writes to your code and never executes it.

// pricing

Free to start. Cheap to love.

Free

£0forever

Security scan, guardrails, skills, and the key vault — on up to 3 repos.

Start free

Pro

most popular

£4.99/mo

Drift verdicts, alerts, unlimited repos, signed records, premium templates and the Warden.

Go Pro

Teams

£22.50/mo from

One bill, every seat on Pro. White-label options and org-wide overview, 5 to 50 seats.

See team plans

Put your agents on the rails.

Connect a repo, compile your contract, and get an independent verdict on the next change your agent makes. Free to start, in under a minute.

Wardrail — the independent trust layer for AI-written code