Map
Python AST and JS/TS analysis build one dependency graph of your code.
01 — AI-powered digital twin for software reliability
Every change
breaks something.
DevDNA shows you what.
DevDNA maps your Python and JavaScript/TypeScript code into a dependency graph, shows the blast radius of a change, and then verifies — or disproves — its predictions with a pytest fault-injection run. Predictions stay PREDICTED until a run verifies them.
02 — Interactive demo · Illustrative only
Click a file.
See its blast radius.
This demo is illustrative, not a real codebase and not a real pytest run. Click any file to light up what it touches. Then run the labelled simulation to see which predictions verify.
Predicted2
SIM Verified—
SIM Disproved—
Files in this illustrative graph
PREDICTED
Blast radius: 2 file(s)
Simulated run only — illustrative, not a real pytest run.
03 — How it works
Map. Predict.
Prove it.
Predict
Change a file and DevDNA lists its blast radius — deterministically, not by guessing.
Verify
A controlled fault-injection run checks each prediction against your pytest suite.
Terminal above types itself on first view and updates with the selected file. It is illustrative — the simulated run is not a real pytest run.
04 — Features
Built for the moment
before the breakage.
Dependency graph
Your codebase as one connected map, across Python and JS/TS.
Blast radius
Exactly which files a change can reach, before you ship it.
Deterministic impact
Impact computed from the graph — no black-box score to trust blindly.
Pytest verification
Predictions are checked by fault injection against your tests.
Predicted vs verified
A prediction stays PREDICTED until a run verifies it. That distinction is the product.
AgentGuard · In Development
Read-only audit of AI-agent permissions, MCP tools, and attack paths. This module is in development, not shipped.
Scope, stated plainly: JS/TS-only projects are analysis-only today — the graph and predictions work, but experiments verify with pytest, so a JS-only repo has nothing to verify against yet.
05 — Module 01 · In Development
AgentGuard
A read-only audit for the agents now touching your code: what permissions they hold, which MCP tools they can call, and which attack paths those permissions open. In development — not a shipped, blocking gate.
06 — Waitlist
Join the waitlist.
0 on the waitlist. That number counts real submissions only.
07 — Testimonials
What people say.
Owner? Unlock delete controls
Enter the OWNER_TOKEN set in Cloudflare Pages (never shown on the public page). It is kept in this browser session only.