Your system already carries its architecture. Let ProvenMap model it.
Software architecture visualization and intelligence, derived from the system itself: ProvenMap builds the living map, then lets anyone on your team ask it questions. Install one plugin, run four commands, and your first board is live.
No credit card · Upgrade when your team grows
The short version · under two minutes
The first ten minutes
Four commands to your first board.
Install the plugin in Claude Code, Codex, or Cursor, open one repository, and run it. No migration, no rollout, no diagramming.
1 · Install
/plugin marketplace add provenmap/pmap-claude
/plugin install pmap-code@provenmapRestart Claude Code so the commands load. Same marketplace for Claude Cowork.
2 · Run, in order
/loginSign in through the browser and bind this repository to a board./analyze-archetypesSettle which kinds of component this codebase has. Required before /analyze./analyzeRead the codebase and build the layered map. Incremental afterwards./syncPush the result to your board.
Lost? /start reads the repository’s state and tells you the one command to run next.
See it working
The product, unedited.
Map · 1:08
Connect a board to the code plugin
One system, one repository, mapped and synced to a board.
Discover · 0:51
What your mapped board can already tell you
/discover finds the insights and context boards worth showing.
Ask · 0:52
Ask ProvenMap, get a context board
A board drawn to answer one question, ready to share.
What leaves your machine
Your repository is never uploaded.
The plugin reads it in your session and sends only the derived description of your architecture, only when a command runs: one you type, or one on a schedule you set.
- Component and connection names, short descriptions capped at about 500 characters, and the longer explanations your agent writes for them
- File paths and line ranges
- The module paths your import statements name, and the symbols imported between files
- Passages /ground quotes from your repository's documents as evidence, each capped at about 500 characters
- What /aspects reads: your API routes, table and column definitions, message channels, page routes, and the full text of authorization policy files
- Screenshots of your running app that /inspect captures, with the page address and on-screen element labels
- Insights, context boards and work-item notes your agent writes about your code
- Findings /monitor draws from your production signals, pushed as insights
- What your agent writes to a workspace through pmap-workspace, including, at /setup, the manifest and dependency names it scanned
- Your repository's git remote URL, branch, current commit and first commit
- The project name, and the repository folder name at sign-in
- A one-sentence summary of a sync, when your agent writes one
- Which agent pushed (Claude Code, Codex or Cursor), the plugin version, the operating system type and a one-way hash of the machine
- Who committed in each period /changes reads, by name only and never by email address, each paired with the coding agent, model and editor their commits name; a short summary of each piece of work; and up to five changed file paths per change rule
- Each scheduled run's outcome, branch and commits, up to five file paths with uncommitted changes, and, when it stops, the error line, capped at 1,000 characters
Credentials stay in a project-local file. Security details · Privacy policy
Start free
Map your architecture. Then ask it anything.
Map your first system on the Free plan. Connect the sources behind your system, get living boards, and give your whole team answers — without the diagramming and without the wait.
1 editor · 5 free viewers · 5 boards · 2 connected sources · 5 published boards · 300 work items · 25 insights · 10 free Chat messages
No credit card · Upgrade when your team grows