WYSTAN.AI
Get started
WYSTAN.AI
How it worksLive scanDocsPricingGet started
ai code governance · feature level

Your agents ship features. Know which — and what they touched.

Wystan reads your git history — no SDK, no instrumentation — and keeps a living, scored map of your product’s features and user flows. Every PR gets a comment naming what it touches, its risk and whose code it is; every Monday, Slack gets what moved.

$99 / month per org · unlimited developers · priced per repo, never per seat

Scan your repo — freeSee a real PR comment
what you actually get

No new dashboard to babysit.

The map lands where your team already works: a risk comment on every pull request, a weekly digest in Slack, context for your coding agent over MCP. The dashboard is there when you want to go deeper.

GitHub PR comment
Slack digest
GitHub PR comment
Slack digest
GitHub PR comment
Slack digest
GitHub PR comment
GitHub PR comment
Slack digest
Slack digest
GitHub PR comment
GitHub PR comment
the problems we solve

The questions your tools
can’t answer.

Five questions, five instruments. Every metric has a public methodology page: what it measures, how it’s computed, and where it’s weak.

Agents write half our code — does anyone know what’s happening in there?

Per product feature: how much of the code is agent-committed — a lower bound, read from git trailers or your git-ai notes — whether it sticks or gets rewritten within a month, and whether a human actually reviewed it.

lower boundAI % · per feature

This PR looks fine. What does it actually touch?

Every pull request annotated with the product features and user journeys it reaches, their validated fix-risk and blast radius — plus a warning when you’re editing code whose author left.

AUC 0.78risk predictor · methodology

Is Checkout getting worse — or just busy?

Health, bug-fix ratio, rework and ownership per feature, on stable identity lines across rescans — trends a diff-scoped reviewer can’t see, because the signal only exists across time. The weekly digest reports movement, and stays silent when there is none.

weeklytrend lines · stable IDs

Tests are green. Why don’t I feel safe?

Coverage in product language: which user journeys are exercised end-to-end, which only by integration tests — and which are bare. Estimated from git history, merged with your lcov when you upload it.

7 signalsjourney coverage · how it’s computed

I just inherited this codebase. What am I standing on?

A human-verified audit: the feature map, bus factor and orphaned code, licences and CVEs joined to the features that import them, and a prioritised risk register — a report you can hand to a board.

48–72 hhuman-verified audit
Every metric is documented — what it measures, how it’s computed, where it’s weak. Engine source-available (FSL).Read the metric docs →Browse 47 real scans →
under the hood

Where those comments come from.

Every risk line above is read off a map — one scan turns your code into a map of your product’s features. Four layers; the portal below is the scan at work — code goes in, product comes out. See it on a real scan →

the product

One scan from git history. A map for your team and precise context for your AI agent.

We read your repository’s history — not a README, not an SDK — and reconstruct the features, the flows inside them, and the files each one touches. Then we score every feature and attach your runtime signal.

01
the code

Your code & infrastructure

Services, gateways, queues, datastores — the real architecture in your repository, exactly as it’s wired together. This is the ground truth we read.

02
wystan

One scan maps code to product

The engine between the code and the product. A scan sweeps the whole repository and reconstructs the map — this is our integration, doing its work.

03
the product

What your users actually get

The product as people use it — the features, and the flows users travel between them.

04
the scan result

A map of your product, scored

Product features and user flows, each scored. Feature names surface over the map; red hotspots flag what needs attention or a refactor; live PostHog traffic is mapped onto features — so you see where the activity, and the risk, really is.

Product featuresUser flowsHotspotsCoverageHealthPostHog traffic
Sentry error→the feature & flow it breaks
PostHog event→the feature & flow it touches
Explore a real feature map →
the product

Product, on top

Each product feature is built from developer features, and the flows are the paths users travel through them.

wystan — the core

The scan, in the middle

The engine is the product — the scan that maps the code below into the product above. Where it strains is where risk concentrates.

the code

Real system, underneath

Services, gateways, queues and datastores — the actual architecture. Every feature is grounded in real code — and the map proves it.

score

Health · risk · churn · coverage

Bug-fix ratio, hotspots, ownership and test coverage — per feature and per flow, so risk stops being invisible.

overlay

Sentry errors · PostHog usage

Live runtime signal mapped onto the map — what’s breaking, what nobody touches — no SDK, no instrumentation.

feed

Context for your AI agent

The same map an engineer reads becomes precise, structured context an agent can query over MCP before it edits a line.

security

Your code stays yours. Here’s exactly what we hold.

Source is read in memory and never written to disk on our side. Every scan is encrypted at rest under a per-scan key wrapped by an HSM-backed KMS key, and decrypted server-side only for authorised reads — every unwrap audit-logged. Feature names and metrics are readable metadata today; Private mode, which encrypts those too, is on the roadmap.

in memory · plaintext, then discardedat rest · ciphertext
Encrypted at rest, per-scan keysAES-256-GCM with a one-shot key per scan, wrapped by an HSM-backed KMS key; decrypted server-side only for authorised reads, every unwrap audit-logged.
No source code stored at restHistory is read in memory, mapped, and the working copy is discarded after each run — on every tier.
Encrypted in transit · KMS-backed secretsTLS everywhere; credentials live in a key-management service, never in code.
Read-only git access, least privilegeWystan never pushes, tags, or writes to your repositories.
Isolated, ephemeral scan jobsEach scan runs in a fresh container with no persistent volume; the working copy is destroyed when the job ends.
Source-available engine (FSL-1.1)The engine that touches your code is published under a source-available licence; every release converts to Apache-2.0 after two years.
Read the security page
get started

Point Wystan at a repo.
Get the map back.

Read the docsScan your repo
read-only git accessno SDK to installfull report after every scan