CartographCompute the architecture. Prove the relationships.

A local-first engine that derives a cross-language architecture graph from source, and shows the evidence behind every edge it draws.

Read the technical writeup

Install and commands

Source on GitHub

Across the stack, not within a language

A TypeScript call, the HTTP route it reaches, the Python handler behind it, the model that handler uses and the table the model maps to — joined by static analysis across languages that share no symbol table.

How the resolver works

Evidence on every edge

Every relationship carries its provenance, a confidence and the evidence for it, and the domain model will not construct an edge without them.

The domain graph

Refuses rather than guesses

An unresolved URL segment stays unknown, an ambiguous match produces no edge, and no language model ever constructs graph structure.

ADR-0007

Measured, failures included

Benchmark results are published with their scope, their denominators and their failures, and every figure the writeup uses is traced to a committed artifact.

What is measured

Local-first

No source code leaves the machine. AI is opt-in per repository and off by default, and the product is fully useful with no API key and no network.

Security

Status

Milestones M00–M16 are accepted; M17, the public launch, is in progress. The current version is 0.1.0. What is and is not implemented is set out in the README, and nothing on this site describes behaviour the code does not have.

Try it

bash

cargo install --git https://github.com/Rexy-5097/cartograph cartograph-cli
cartograph .

Then follow one relationship across the stack with cartograph trace <symbol>. The v0.1.0 release notes cover installation, and the command-line reference documents every command, including blast and diff.

What ships today

Surface	What it answers
The CLI — `cartograph .`, `trace`, `blast`, `diff`	What a repository contains, how one relationship crosses the stack, what depends on an artefact, and what a change did to the architecture
MAP, the desktop application	The architecture graph, drawn, with the evidence behind any edge
`cartograph-mcp`	The same graph, queried by an MCP client over one repository granted at launch
A pull-request review workflow	`cartograph diff --markdown` posted as a comment — configured for this repository's own pull requests
ASK, in the desktop application	An explanation of already-derived evidence in which every claim must cite it verbatim — opt-in, and with no in-product way to supply a key yet, so in practice it shows the evidence unexplained

TypeScript, TSX and Python only, within a declared supported subset; outside it Cartograph refuses rather than guesses. The resolver lists what is implemented and what is not.

Where to start

The technical writeup — resolver design, benchmark results including failures, and calibration, with every figure traced to a committed artifact.
Benchmarks — what was measured, what has been published, and why Gate 2 remains unevaluated.
Architecture decisions — the reasoning behind each boundary, recorded as it was made.