Architecture map
A local scan groups your files into features and foundations. Open any component to see its modules, functions, classes and API routes with file paths and line numbers. Press / to search them all.
Desktop workspace for Claude Code and Codex
Your codebase, mapped.
Your agents, tracked.
Kalvyr scans your project into a map of components, then shows where your coding agents are reading, editing and running checks, as it happens. Preview the running app and chat with your agents in the same window.
open-field/Architecture
Replay · example data
Watching files
2 sessions
0 events
Select a component to look inside
- Reading
- Editing
- Running a tool
- Running checks
- Thinking
- Waiting
- Dashed = inferred location
- Agents
- Claude Code, Codex
- Setup
- No hooks or agent settings to change
- Runs on
- Your machine, on
127.0.0.1 - Accounts
- Your own agent logins, no API keys
What you get
Everything between your agents and your code, in one window.
Panels below are illustrations with example data.
Agent tracking
Kalvyr follows the sessions your agents already write, whether they run in a terminal or in the app, and places each step on the map.
Plan tracking
When an agent makes a plan, it restates it as steps tied to your components. Progress comes from its task list and the edits Kalvyr observes.
Live preview Pro
Run your app beside the map. Vite apps update live, and Follow agents outlines the elements an agent is working on. Next.js, Create React App, Express and Expo web run through their own dev script. Plain HTML sites are served read-only.
Component previews
Render exported React components with no running server, at desktop or 390px width. Read database tables from SQL and Prisma, and API routes from your Express, Fastify or router calls, straight from source.
Catch up and history
Come back to a recap by component: files changed, connections added, checks that ran. Step through snapshots of how the architecture changed.
Agent chat Pro
Send work to Claude Code or Codex from the app, with images. Pick the model, effort and permission mode, and approve each tool request yourself.
Plan limits at a glance
A ring in the chat box shows the limit closest to running out. Click it for each limit and when it resets, read from the agent's own CLI. Checking uses none of your plan.
AI descriptions
Connect a project and your agent names the components and explains each one in plain language. In Pro, it can update just the parts that changed after agent work, using your plan.
How it works
Connect a folder. Keep working the way you do.
-
01
Connect your project
Choose a local folder. Kalvyr scans it on your computer and draws the map, then your agent describes it. Starting from nothing? Describe the project and get a planned architecture instead.
Connect & scan → D:\code\open-field -
02
Run your agents
Use Claude Code or Codex wherever you like. Sessions working inside the folder are picked up automatically, with nothing to install in the agent.
claude · codex · or chat in Kalvyr Pro -
03
See what happened
Watch the work move across the map, check the result in the preview, and review what changed by component instead of file by file.
Since you last looked → Mark reviewed
Local-first
Your project stays on your computer.
The map, the scan and session tracking all run in a small service on your machine. Kalvyr never uploads your code. When an agent describes a newly connected project, or you send a chat or ask for AI analysis, your agent's own CLI talks to its provider, using your account.
Your machine
Kalvyr window
↕
Local service 127.0.0.1
↕↕
Project filesread locally
Agent CLIsyour login
↓only on your request
Claude / OpenAI
It labels what it knows, and how it knows it.
Code detectedSource files were found. It says nothing about whether they work.
Reading · inferredThe location was worked out, not named by the agent. Drawn with a dashed outline.
Build passedShown with its source and time. Later edits mark it outdated.
Completion reportedAn agent said it's done. Test and build results stay separate.
Free & Pro
Start free. Move into Pro when you build here.
Pro is $9/month, billed through Stripe. Cancel any time.
Pro
$9/month
Everything in Free, plus agent chat, live preview and saved conversations.
Get Pro · $9/month
Feature
Free
Pro $9/month
Architecture map, search and component detailsIncludedIncluded
Agent tracking, plans, activity and historyIncludedIncluded
Component previewsIncludedIncluded
Live preview of your running appNot includedIncluded
Chat with Claude Code and CodexNot includedFull pane beside map and preview
AI descriptions of your componentsOn connect and on requestCan update after agent work
Auto-detect agentIncludedIncluded
Saved conversations you can resumeNot includedSaved and searchable
Project list and session managementNot includedIncluded
Create a Kalvyr account and subscribe on this site, then choose Sign in in Kalvyr and confirm in your browser. No license key. Payment runs on Stripe. Agent usage always comes from your own Claude or OpenAI account and is subject to its limits.
Download
Get Kalvyr for your desktop.
Kalvyr is available for Windows and Linux. Choose your package below.
- Node.js isn't needed: the installer includes its runtime
- Claude Code or Codex CLI, for AI descriptions and Pro chat
- Early builds are unsigned, and updates are manual
FAQ
Good to know.
How much is Pro?
$9 a month, billed through Stripe. Choose Get Pro above or in Kalvyr, create your account and subscribe. Then choose Sign in in Kalvyr and confirm in your browser: Pro turns on, with no license key to copy. Cancel any time from your account page or Manage subscription in Kalvyr: Pro keeps working until the end of the month you paid for. Free stays free.
Do I need API keys?
No. Kalvyr uses the native Claude Code and Codex CLIs with your own account. Install the tool you want and sign in. Pro can also connect an account through your browser.
Does my code leave my computer?
Scanning, the map and session tracking run locally, and Kalvyr doesn't upload your code. The first time you connect a project, Claude Code or Codex reads it to describe its parts, if one is available. That run, and any chat or AI analysis you ask for, may send project content to the agent's provider using your account.
Do I have to change how I run my agents?
No. Kalvyr follows Claude Code and Codex sessions that work inside the connected folder, with no hooks or settings changes. Sessions you started elsewhere are never written to: sending a message from Kalvyr starts its own conversation.
Can an agent change my files from the app?
Yes, in Pro. Chat runs a real agent with its own permissions. Tool requests are shown for you to allow or decline unless you choose a more permissive mode. Stopping a run doesn't undo edits already made.
Which apps can Live preview run?
Live preview is part of Pro. It runs Vite apps, apps with a dev, develop, start or serve script (such as Next.js, Create React App, Express and Expo web), and plain HTML sites. Live preview runs your project's own development code. It doesn't install dependencies or start separate backend services. Native mobile apps, Electron apps and server-side templates have no page to preview.
What's available to download?
Kalvyr is available for Windows x64 as an installer, and for Linux x64 as an AppImage, Debian package or .tar.gz archive. Downloads and SHA-256 checksums are hosted on GitHub. The Windows installer is currently unsigned. Once installed, the Windows app, AppImage and Debian package update themselves.
Does the map prove my code works?
No. The map describes code structure and observed agent activity. AI descriptions are interpretations, and reported checks are shown as evidence with their source, not as a guarantee.