vibesequence

Website · How it works · Benchmark

Turn your codebase into zoomable sequence diagrams that show when a user does X, who calls whom,
and in what order. Use them as a map while you build, review or learn the code, on your own or with an
AI assistant.

Inspired by graphify, which turns a codebase into a knowledge
graph so an AI assistant can navigate it without re-reading every file. vibesequence takes the same idea
and adds what a graph leaves out: the order of calls and the branches around them (if, loops, error
handling), drawn as user sequence diagrams that you and the assistant can both read.

Why

When an AI assistant edits your code, it is hard to see what actually changed in the behavior.
vibesequence reads your code (no running it), finds the entry points (HTTP routes, CLI commands, handlers),
and draws each one as a sequence diagram:

• Zoom between package, module, class and function level. Click a blue call to open it on its own.
• Play a flow arrow by arrow, with the file:line of each step.
• Live: while the AI edits, changed steps turn green and you can play only the changes.
• AI friendly: the same flow is available as compact text, so your assistant reads one diagram instead of eight files.

Supports Python (stdlib only), TypeScript / JavaScript (Next.js, Express) and Go (net/http, Gin, Echo, Chi, Fiber, Cobra).

Does it save tokens?

We benchmarked Claude Code on the same tasks with and without vibesequence. Every run passed in both setups.

Impact questions save the most: one vibesequence callers call replaced 6–7 greps and reads (−53% tokens).
For edits, the agent still has to read the files it changes, so on the tiny example projects the map is
extra overhead. Full method, per-task results and caveats: docs/benchmark.

Install

From PyPI:

pip install vibesequence
pip install "vibesequence[ts]"   # with TypeScript / JavaScript support
pip install "vibesequence[go]"   # with Go support

Quick start

vibesequence build .        # parse the repo into vibesequence-out/
vibesequence serve          # open the viewer at http://127.0.0.1:8765

Try it on the bundled examples: clone this repo and run vibesequence build examples.

Commands

Viewer shortcuts

Space play · ← → step · [ ] level · Z zen · D dark mode · \ sidebar · ⌘+scroll zoom · drag to pan

Roadmap

• Go: tree-sitter extractor, entry points for net/http, Gin, Echo, Chi, Fiber and Cobra commands
• Java: Spring controllers
• Smarter resolution: return-type inference, dependency injection, interface → implementations
• Diff against a git ref: review a PR as sequence changes, not line changes
• MCP server: let the assistant call list_flows / show_flow instead of reading files
• Runtime traces (optional): mark the paths that actually ran

Contributing

Open for contributions! Pick a roadmap item, fix a bug, or improve the docs. Issues and pull requests are welcome.

• New language: add one extractor class in vibesequence/extractors/ that outputs ModuleFacts
  (see extractors/base.py). The resolver, viewer and CLI work without changes.
• Set up: clone the repo and run pip install -e ".[ts,go]".
• Run the tests before opening a PR: python tests/test_flow.py
• Not sure where to start? Open an issue and say what you want to build.

Learn more

How the engine turns code into sequence diagrams (extract → resolve → expand → project → render), plus
live mode, editor links and the AI hook: docs/how-it-works.md.