An AI writes code fast: 548 files and 8,373 symbols, too much to hold in your head.
Your editor answers about one symbol at a time. Code Helper sees the whole project, across 13 languages.
It indexes 13,128 references in 1.7 seconds and draws how the files connect.
The real app, running on a real project.
Your AI assistant asks it too: what uses OrderingContext.cs comes back from the index.
Ambih Code Helper is free on one project, on Windows, with no account needed.
What it actually does
grep finds text. This finds structure. Four things follow from that.
A map, not a folder tree
Code Helper counts every file, symbol and reference in your project, sorts each file by the job it does, and draws how a request travels from the user down to the database. You start from the picture, not from the folders.
The whole project counted the moment it is indexed
Every file sorted into UI, controllers, services and entities, by what it does, not where it sits
The path from the user to the database, drawn for you
What the video shows
The overview of a real project, dotnet/eShop: 570 files, 3,452 symbols, 9,901 references.
The files sorted by what they do: UI, controllers, services, entities.
Then the architecture map: from the user, through the controllers and services, to the database.
Your whole project on one canvas
Not one file and its neighbours — the entire project, drawn from a real parse. Every file placed by the role it plays, every dependency that genuinely exists, and the boundaries where your code meets a database, an API or a queue. Unveil one layer at a time instead of six hundred boxes at once.
The whole system first, then drill in — not one file at a time
Hover any file: orange for what it depends on, blue for what depends on it
Open one role group at a time — services, then controllers — to keep it readable
Split a monorepo into the apps it actually deploys as (Pro)
What the video shows
Every file of a project drawn at once is a tangle nobody can read.
Code Helper places each file by the role it plays, beside the boundaries where the code meets a database, an API or a queue.
Then it shows one layer at a time.
Keep your bearings in a project that grew
Every file is sorted by what it does — a controller, a service, an entity, a screen — worked out from the code itself, never from folder names. On a codebase you have maintained for years, that is the difference between roughly knowing where things live and being able to point at them.
See where your code touches the outside: the database, the API, the queue
Every label says why it was given: declared by the framework, or worked out
Click any box to open the file, its neighbours and its code
Right on a project that ignores convention, because folders were never the evidence
Compare two commits by structure, not by lines
A line diff tells you which text changed. This tells you what the change did to the project: which functions appeared or vanished, which files started depending on which, and which parts of the app it actually reached.
A picture of the change, not a wall of red and green
Connections added and removed are drawn in, not left to be inferred
Full before, after and inline diff for every change
What the video shows
A line diff of a real commit in dotnet/eShop: 12 files, 99 lines added, 189 removed.
Code Helper shows what the change did: OrderQueries stopped depending on NpgsqlDataSource, lost MapOrderItems(), and now depends on OrderingContext.
VS Code extension — in development
The same map inside VS Code
Right-click any file in VS Code and see what connects to it, without leaving the editor. Your AI assistant can ask the same questions over MCP, so it looks things up instead of guessing. App, editor and assistant read one index, so they can never tell you different things about the same project.
Right-click a file or symbol in VS Code → Show connections
Callers on the left, what it uses on the right. Drag the boxes around
31 tools your AI assistant can call: definitions, callers, dependencies, structure
One shared index. Nothing is analysed twice, and nothing disagrees
What the video shows
Right-click a file in VS Code and choose Show connections.
The files that call it appear on the left, what it uses on the right.
Your AI assistant asks the same question over MCP and gets the same answer, from the same index.
Parses, with real language grammars:
C#
JavaScript
TypeScript
Python
Java
Kotlin
Dart
PHP
C++
Rust
Go
Ruby
Scala
AI
Answers only a model can give
Parsing tells you what your code is. A model can tell you what it is for — so Code Helper asks the agent already on your machine, and keeps the answer.
Your own agent, no key to paste
Code Helper runs the AI agent already installed on your machine. There is no API key to paste, no second subscription, and nothing goes to us.
What this application is
It reads the routes, pages and entities your project declares, then names it in a sentence — “an online shop with cart and checkout” — and shows the evidence it used.
Grouped by what it does
The model groups your domain nouns into capabilities — Checkout, Sign-in, Cart. The dependency graph then finds the files each one reaches, so every file keeps a reason you can check.
Your findings, explained
The Improve tab measures; the model says what a finding means for this application and which to look at first. The measurements do not move, and the two are never mixed together on screen.
Using this? Other developers would find your experience useful.
Write a review
Questions people ask
I do not use AI to write code. Is this still for me?
Yes, and that is the side it was built for first. The map, the file labels and the commit view are things you look at. Most of what Code Helper does is answer questions a person has about a project: what is this file, who calls it, what breaks if I change it. An AI assistant can ask the same questions, and that is genuinely useful, but nothing here needs you to be using one.
I already know my codebase. What would I get out of it?
Three things nobody holds in their head. What has quietly stopped being used. What a commit actually rewired, as opposed to what it touched. And which files everything else leans on, so you know what is expensive to change. There is also the version of you in six months, and whoever you hand the project to: the map is already drawn for them.
My project is old and messy. Will it actually work?
That is the case it was built for. Labels come from the code itself, never from folder names, so a project that ignores every convention still maps correctly. It reads what is actually there. A tidy project is not a requirement, it is just a project that was already easy to read.
How much work is it to get something useful?
Point it at a folder. There is nothing to configure, no project file to write, and nothing to add to your repository. Indexing this app’s own source, 548 files, takes about 1.7 seconds, and the map is on screen when it finishes.
How is this different from "Find All References" in my editor?
Your editor answers about one symbol at a time, in one language, while the project is open. Code Helper keeps a map of the whole project across 13 languages. It answers what your editor cannot even be asked: what shape is this app, which files depend on which, and what did that commit really change. Your AI assistant can ask the same things.
Does my code leave my machine?
No. Indexing and every query run locally, and the index is a SQLite file on your own disk. There is no account required to use the app and no code is uploaded anywhere.
Which languages does it cover?
C#, JavaScript, TypeScript, Python, Java, Kotlin, Dart, PHP, C++, Rust, Go, Ruby and Scala. It reads them with the same parsers your editor uses, Roslyn and tree-sitter, so it is reading your code the way a compiler would rather than guessing at it. Templates count too: Blade, Twig, XAML and Vue views link back to the code behind them.
Is the dead-code report safe to act on?
Treat it as a shortlist, not a verdict. Every finding carries a confidence level, and the report says how much of the project it could answer for. Frameworks call code in ways no tool can see from the outside, so this one deliberately under-reports rather than telling you to delete something that is live.
When does it ship, and what will it cost?
There is no date yet. It will be free on one project, $99 a year for unlimited projects, or $219 once with twelve months of updates. Code Helper is priced on its own, so an Ambih Monitor subscription does not include it and buying one does not include the other.
Pick your platform
Windows is here today. macOS is in development — tell me you want it and you will hear the day it lands.
