Early prototype · mapping real repositories today
See your codebase as a living city.
Neonmap turns a repository into an interactive 3D map: modules become districts, files become towers, and imports become glowing roads. Next, Claude adds the understanding: what each file does, and answers to your questions.
● Real data · github.com/honojs/hono · commit f78a245
- 312
- files
- 11
- modules
- 833
- imports
- 2,878
- commits
The problem
Large codebases are cities without maps.
Onboarding takes weeks
New engineers spend their first month just finding their way around.
Dependencies tangle silently
A small change in one module breaks three teams you didn't know depended on it.
Docs drift, code doesn't
Architecture diagrams go stale. The code is the only source of truth, and nobody can see it.
Handovers lose knowledge
When a project moves between teams or vendors, the mental map leaves with the people.
The map
From repo to city in minutes.
The analyzer resolves every import, measures every file and reads the whole git history. This city is honojs/hono, an open-source web framework, mapped by the Neonmap prototype. Every tower is a real file.
- District
- a module or package
- Tower
- a file; height is lines of code
- Windows
- lit by commit activity
- Arcs
- imports between modules
Time-lapse
Watch your architecture grow.
+15 files, 312 in total.
- –
- files
- –
- lines of code
- –
- modules
Blast radius
Know what breaks before you merge.
src/context.ts
181 files in 9 modules could break.
- middleware/60
- adapter/51
- helper/35
- src/11
- client/7
Click any tower in the city, or try one of these:
Ask the city
Ask questions. Get flown to the answer.
- Complexity heatmap working
- Hotspots glow red, so you know where to refactor first.
- Blast radius working
- See every file a change could break before you merge.
- Answers from Claude in development
- Ask anything in plain language; answers cite the files they come from.
- Always current planned
- The map updates on every push. No diagrams to maintain.
Ask the city
Real dataFlying to client/
client/ has the highest average complexity. Its biggest hotspots are client/types.ts (about 76 decision points in 353 lines) and client/client.ts (about 56 decision points in 217 lines).
These answers come from the import graph and git history. Free-form questions answered by Claude are in development.
Under the hood
Deterministic analysis first. Claude where understanding matters.
- 01working
Parse
Every file is read and its imports resolved. TypeScript and JavaScript today.
import resolution · tree-sitter next
- 02working
Graph
Deterministic analysis builds the dependency graph, complexity estimates and blast radius. No AI guessing.
import graph · metrics
- 03working
History
One pass over git log gives every file's age and how often it changes.
git
- 04working
Render
The graph is laid out as a city and sent to your browser as a small JSON file.
three.js · WebGL
- 05in development
Understand
Claude reads each file and module and writes summaries, responsibilities and risks. Only changed files are re-read.
Claude · Batch API · prompt caching
- 06in development
Ask
Claude answers questions with tools over the graph: search, open a file, trace a dependency. Answers cite files.
Claude · tool use
Steps 01–04 produced the city on this page from real data.
Roadmap
Next: where you already work.
GitHub App
nextConnect a repo with read-only access, then get a review comment on every pull request: blast radius and a link to fly to the change.
Claude Code (MCP)
nextAn MCP server gives Claude Code the map, so it can check who imports a file and what could break before it edits.
VS Code
laterReveal the current file in the city, and see who imports it without leaving your editor.
Slack
laterA weekly architecture digest: new hotspots, growing modules, and risky merges.
Built for
Onboarding
New hires explore the city on day one instead of reading stale wikis.
Project handover
Hand a codebase between teams or vendors with the map included.
Legacy modernization
Find the seams to split a monolith, ordered by risk.
Technical due diligence
Size up an unfamiliar codebase in an afternoon.
Questions
Does it change my code?
No. Neonmap only needs read access. It never pushes commits; pull request comments are optional.
Is my code used to train AI models?
No. Claude analysis runs on Anthropic's commercial API, and Anthropic doesn't train models on API inputs or outputs by default. A self-hosted option is planned for teams that need code to stay in their own cloud.
Which languages are supported?
The prototype maps TypeScript and JavaScript today. Python, Java, Go and C# are next, using tree-sitter grammars.
How is this different from a dependency graph?
Graphs turn into hairballs after a few hundred nodes. A city keeps a stable layout, so you build spatial memory: you remember where things live, like streets in your own town.
What does it cost?
Free during the beta. After that, paid plans per repository, with a free tier for open source.
Why is the demo honojs/hono?
It's a well-known open-source TypeScript project with clear modules and years of history, so you can check the map against the real code. It's MIT-licensed and not affiliated with Neonmap.
Early access
Map your first repo.
We're looking for a few engineering teams to map their repositories first. Join the waitlist, or email us if you'd like your repo mapped during the beta.
Questions? admin@neonmap.tech
Who's building this
- LouisCo-founder
- TienCo-founder
We're building tools that make large codebases understandable, starting with this map.