AnaxiGraphGitHub

Open-source architecture intelligence

Understand the system.
Guide the agent.

AnaxiGraph turns a fast-growing codebase into shared architectural context—so people and AI coding agents can build without creating spaghetti code.

Run it locally Explore on GitHub
  • Local-first
  • Read-only scans
  • MCP-native
  • Apache 2.0
Chaotic repository

Code accumulates faster than a person—or a coding agent—can keep the whole system in context.

AnaxiGraph intelligence engine
AnaxiGraph
  • Structural mapping
  • Semantic mapping
  • Dependency graph
  • Capability model
  • Fresh Eyes review
  • AI comparison
  • Change tracking
  • Impact analysis
Architectural clarity

A shared, evidence-backed model explains the system and guides the next change.

Keep the speed of AI coding.

Recover the architectural judgment it tends to lose.

The architecture sidekick

Not another code search.
A second opinion on the whole system.

01System understanding

A map that means something

See capabilities, areas, modules, symbols, dependencies, ownership boundaries, and how they evolved—not just a cloud of nodes.

02Agent guidance

Context your agent can use

Ask where a feature belongs, what it may affect, which tests matter, and which patterns already exist before another change is written.

03Architecture oversight

Pressure against entropy

Surface boundary drift, cycles, hotspots, duplication, oversized modules, dead-code candidates, and refactoring opportunities with evidence.

One model, two interfaces

The human and the agent see the same architecture.

The dashboard makes a codebase legible. MCP gives that same evidence to the coding agent already doing the work. Both operate on the AnaxiIndex instead of rebuilding a partial mental model in every chat.

Facts stay facts.Static evidence, semantic interpretations, and recommendations remain distinguishable—so useful guidance does not pretend to be certainty.
1

Map

A deterministic, read-only scan records files, symbols, relationships, history, churn, and architecture evidence.

2

Understand

Your coding agent uses its own tokens to describe module intent and relationships. Hashes prevent unchanged code from being reread.

3

Guide

Before a change, retrieve the smallest useful context, existing patterns, blast radius, tests, and architectural options.

Working in minutes

Bring the architecture sidekick into any repository.

Run the dashboard, enable agent-funded semantic understanding, and connect Codex in one local command.

Read the quick start View on PyPI
terminal
$ uvx anaxigraph up . \
  --open \
  --semantic agent \
  --connect codex

Scans locally. Target code is never executed or modified.

Software should become easier to understand as it grows.

Keep the architecture coherent.

Build with AnaxiGraph