About
Software drifts away from the drawing. That is the problem.
Graphlit exists because the map and the territory always diverge, and nobody notices until something breaks. We keep them provably in step.
Why we built this
The diagram was wrong and everybody knew
Every team we have worked on had an architecture diagram somewhere: a whiteboard photo, a stale draw.io file, a slide from a review two years ago. Every one of them was wrong, everybody knew it was wrong, and nobody could say where it was wrong without reading the code.
The usual answer is discipline: update the diagram with every change. That has never worked anywhere, because a diagram that has to be maintained by hand is a second codebase with no tests and no compiler.
So the drawing needs to be checkable. Not prettier, not automatic. Checkable. If every box maps to real files and every arrow maps to a real call, then a machine can answer “is this still true?” on every commit, and the answer is a fact rather than an opinion.
That is the whole product. You draw the system, we turn it into a typed graph, and from then on we can prove, commit after commit, whether the code still matches. Building the app from the drawing came second, and it is the part we consider the easy half.
Who we are
A small independent team
Graphlit is built by a small team in India and is entirely bootstrapped: no venture capital, no outside investors, no board. It is early, it is being built in the open, and the people writing it are the people who answer your email.
Being bootstrapped is the reason the refusals below are affordable. Nobody is waiting on a growth curve that would make training on your code, or shipping a feature we do not believe works, the sensible move.
We have not put photographs, headcounts or customer logos on this page, and that is deliberate. We would rather show nothing than a number chosen to look impressive. What we can point at instead is the work: the build log is updated as things land and lists what is broken alongside what is finished.
Email reaches a person, not a queue. prin@cipher.academy. Usually within one working day.
What we believe
Five opinions the product is built on
A drawing that stops being true is worse than no drawing
Every architecture diagram is accurate on the day it is drawn and decays from then on. People stop trusting the diagram, then stop updating it, then stop drawing them. The decay is the problem. Not the drawing.
Generating code is the easy half
Plenty of tools turn a prompt into an application. Almost none can tell you, three weeks and forty commits later, whether the thing they built still matches what you asked for. We would rather own the boring half that keeps working.
Verification should not guess
Drift detection, health checks and contract enforcement are deterministic here: hashes and parsers, not a model's opinion. A checker that is sometimes wrong is a checker you learn to ignore, and an ignored checker is worse than none.
Your code should not have to leave the building
The scanner is deterministic and runs locally, so a graph can be built from a private repository without the source going anywhere. Source reaches our machines only when you explicitly ask us to run an agent on it.
Say what does not work yet
The build log lists what is broken and what is missing, publicly, including the parts that are embarrassing. It is a better filter than a feature list: people who need the missing piece find out now rather than after paying.
What we will not do
The refusals
Worth writing down, because they are the promises that are expensive to keep.
- We do not train models on your code.
- We do not fire an agent at your repository because you drew something. Dispatch is always an explicit action.
- We do not publish review or rating markup we have not earned, and there are no testimonials on this site because there are no customers to quote yet.
- We do not silently delete anything the model no longer sees. Content that vanishes from a drawing is flagged for a human, never removed.
Point it at a repository you already have
The fastest way to judge any of this is to import something real and see whether the graph it derives matches the system in your head.
Free to start. Import a repo and the graph is there in minutes.