See what your agent changed before you merge it.
Changemap maps your real system from the code, shows every change on that map, and makes fixes on their own branch so you can undo them. Use a local model, your own key or your own Claude Code. No credits.

Agents write the code now. Someone still has to understand it.
Almost everyone uses AI to write code. Few trust what it writes, and the time saved writing goes into reviewing.
One map of your system, and everything checked against it.
A map built from the code
Apps, APIs, databases and outside services from your deploy config and imports. Zoom in through packages and folders down to files. Every box says how it was found.
Problems from the start
Import cycles, broken imports, undeclared dependencies, unused files and packages, committed secrets, exposed ports. Facts, each pointing at the place on the map.
Review any agent's work
Changed files are marked on every level of the map. Compare with the last commit side by side, stage, commit, or set changes aside in a stash you can get back.
Fixes you can undo
Every fix is previewed as a diff and committed on its own branch. Checked before you see it: it parses, and no new broken imports or cycles. Merge it, or undo it in one click.
Framework upgrades, done for you
Expo SDK, bare React Native and .NET upgrades, with matching package versions, peer checks and a build check, on a branch like every other fix.
Ask about the selected box
The chat knows what you're looking at. Its edits come back as review cards; commands it wants to run wait for your Run click. An AI review finds bottlenecks and links each one to the map.
Catches weakened tests
Skipped tests, deleted test files, removed assertions, rewritten snapshots, a test script that can no longer fail. Flagged before you accept the change.
Riskiest files first
Every review and the Changes list put the files that need a close look at the top, with the reason: weakened tests, secrets, auth, many dependents.
From a plan to code
Describe what you're building. The model asks the essentials, proposes a plan drawn as a diagram, then builds it step by step on a branch, a commit per step, pausing for anything risky.
Diagrams drawn by an AI are one more thing to doubt.
Changemap reads your repository like a compiler would. When the code changes, the map changes.
Diagrams written by a model
- Guessed from file names, READMEs and samples
- Can show parts that don't exist
- Another source of truth to keep in sync
- Usually in the cloud, with your code uploaded
Changemap's map
- Built from real imports, project files and deploy config
- Every part and link carries its evidence
- Updates when files change or you commit
- Runs on your Mac; the code stays there
Keep your agent. Add the map.
Claude Code in your terminal
Connect it once. It can show its changes in Changemap, open files at a line, point at boxes on the map and ask the map about your architecture.
Your model, no credits
A local model in LM Studio or Ollama, your own Anthropic or OpenAI-compatible key, or your installed Claude Code. Every answer shows its tokens and estimated cost. Changemap adds no markup and sells no tokens.
Your stack
JavaScript and TypeScript monorepos (pnpm, npm, yarn workspaces), C# and .NET solutions. Deploy config from Docker Compose, Cloud Build, Firebase, EAS, nginx and Terraform.
Free while in beta. Simple after.
Changemap never resells AI usage, so the price stays low and predictable.
Free
For understanding any repository
- System and Code maps
- Findings
- Changes view and editor
- Chat with any model
- Claude Code terminal link
- Export to PNG, SVG, Mermaid
Pro
For shipping with agents every day
- Everything in Free
- Fixes on their own branch, with merge and undo
- Fixes written by your model, checked before you see them
- Framework upgrades
- Build a new project from a plan
- AI architecture review
- Keep the version you paid for if you stop
Team
Coming later
- Everything in Pro
- A check in CI that shows what each pull request did to the architecture
- Shared settings and central billing
Prices are planned and may change before launch. Free for open-source maintainers and students.
Before you ask
Does my code leave my Mac?
The map, findings and fixes run locally. Code goes only to the model you choose, and with a local model nothing leaves at all. Changemap's own model tools refuse .env files, keys and credentials.
Do I need Claude Code?
No. It's one option. A local model or your own API key work too.
Which languages?
JavaScript and TypeScript, and C#. Other languages get the system view and version checks for now. Python is planned.
Windows or Linux?
Not yet. The beta is for Macs with Apple silicon.
Will it change my code without asking?
No. Fixes go on a separate branch, chat edits wait for your review, and every command needs your Run click.
Is it another editor I have to switch to?
No. Keep your editor and your agent. Changemap is where you see the system and check what changed.
Try it on your own repository.
We're inviting a small group of developers who work with coding agents. Leave your email and we'll send an invite when there's a place.
Already invited? Your download and beta key are on the personal page linked in your invite email.