Knowledge graph + governed AI agents

See what a change
will break. Then fix it safely.

SystemDNA maps every file, type and field in your code. Rename one field and it shows every file that must change, in the right order. Then a crew of agents fixes each file, checks the work and opens a draft pull request for you to review.

Replaying a real change on marketplace-dashboardLive 3D
  1. ChangeRename successRate
  2. Impact6 files in 2 waves
  3. FixOne agent per file, checked
  4. ReviewA draft pull request opens
0Files mapped
0Lines of code
0Components found
0Dependencies traced
0 msTo work out the impact

Real numbers from the sample repo marketplace-dashboard, scanned by SystemDNA.

The problem

Find and replace guesses. SystemDNA knows.

Renaming a field sounds simple. But text search cannot tell two fields with the same name apart, and it cannot see files that use the field without spelling it out. So you either break something or check every file by hand.

Search for “successRate”Find and replace
2 mistakes
  • .cursor/plans/marketplace_dashboard_prd_b814bbac.plan.mdFound
  • app/products/[id]/page.tsxFound
  • components/ProductTable.tsxDifferent field
  • components/views/ProductsView.tsxDifferent field
  • data/seed.tsFound
  • docs/PRD.mdFound
  • lib/data.tsFound
  • lib/types.tsFound

It gets 6 files right, changes 2 files it should not touch.

Rename Deployment.successRateSystemDNA
Exact
  • lib/types.tsWave 1 · Types
  • .cursor/plans/marketplace_dashboard_prd_b814bbac.plan.mdWave 2 · Tests and docs
  • app/products/[id]/page.tsxWave 2 · Pages
  • data/seed.tsWave 2 · Data
  • docs/PRD.mdWave 2 · Tests and docs
  • lib/data.tsWave 2 · Logic

6 files, fixed in 2 waves so each file changes only after the files it depends on. It also shows the business areas affected: Product detail.

How it works

From repository to reviewed pull request in four steps

01

Connect

Paste a public Git URL or upload a .zip. SystemDNA reads the code. It never installs, builds or runs it.

02

Map

The TypeScript compiler finds every type, field, function, page and doc, and every link between them.

03

Predict

Pick a field and a new name. See every file that must change, the risk, the order and the business areas it touches.

04

Fix

One agent per file, wave by wave, inside strict rules. Checks run on every file, then a draft pull request opens.

The multi-agent run is a preview today. The GitHub agent that makes the edit and opens the pull request is live.

Inside the dashboard

A control room for your whole codebase

Five screens take you from “what do we have?” to “it is fixed and ready to review”. Everything below uses the real sample repo.

systemdna.app/overview
Overviewmarketplace-dashboard
Components
202
Dependencies
643
Files
40
Layers
7
Components per layer
Types
Data
Logic
Components
Pages
Business use
Tests and docs
How links were found
100%
643 by the compilerTypeScript language service, not text search.
The knowledge graph

Your system's DNA, read by the compiler

Every component is a bead. Every link is a rung. SystemDNA uses the TypeScript language service, the same engine behind “find all references” in your editor, so two fields with the same name are never mixed up.

  • Fields103
  • Functions27
  • Types24
  • Components15
  • Pages10
  • Datasets9
  • Business uses8
  • Constants3
  • Docs3
Types and fieldsInterfaces, classes and their fields
CodeFunctions, constants, datasets
ProductPages, components, business uses, docs
What you get

Everything you need to change code with confidence

Fixed in the right order

Files are grouped into waves by what depends on what. A wave starts only when the one before it passes its checks, so nothing is built on a broken base.

Wave 1
types.ts
Wave 2
marketplace_dashboard_prd_b814bbac.plan.mdpage.tsxseed.tsPRD.mddata.ts

A risk score you can read

From 0 to 100, based on how far a change spreads, how critical the code is, personal data and missing tests.

71/ 100 for this change

Three ways to look

A dependency map, a 3D city with one building per file, and a force graph of every node. Your selection follows you across all three.

Map3D cityGraph

Business impact, not just files

See which products and teams a change reaches, so the right people hear about it before it ships.

Product detail

A real pull request

The GitHub agent makes the rename with the compiler, checks that no file gains a type error, shows you the diff, and opens a draft pull request once you say yes.

Safety

Agents that follow the rules of the city

Automation is only useful if you can trust it. Every agent in SystemDNA works inside city laws that are checked on every single action.

Reads code, never runs it

No install, no build, no scripts from your repo. Temporary copies are always deleted.

Permits for every agent

Each agent may edit only the files on its permit. No two agents ever work on the same file.

People approve risky files

Protected files, like shared contracts and personal data, wait for a person before any edit.

Drafts only

Pull requests are always drafts on a new branch. The default branch is never touched.

Your token stays on the server

It is sent as a header only. It never appears in a URL, a log or an error message.

Every action is recorded

Permits, tool calls, blocks, approvals and checks all land in the audit log.

Map your own system in minutes

Connect a public repository or upload a .zip. You get the knowledge graph, the 3D city and a real impact report straight away.