Living architecture · humans + agents

The architecture model your team and your AI agents actually trust.

Point Archmantic at a repo and it reverse-engineers one grounded model — a capability map, diagrams, a data model, and the API surface. Your team reads it as living docs; your agents query it over MCP. Every element traced to a line of code.

View on GitHub

$ npx archmantic analyze && archmantic push

archmantic · payments-api

Capability map

Charge a card● high

Refund payment● high

Dispatch webhook● medium

Reconcile ledger● medium

Context

Polyglot — TS/JS & PHP/Laravel · monorepos · APIs & data models

TypeScript

JavaScript

PHP

Next.js

React

Vue

NestJS

Express

Laravel

Inertia

Livewire

Prisma

Drizzle

GraphQL

PostgreSQL

MySQL

SQLite

One model, two audiences

The same grounded model — read as docs by people, queried as context by agents.

No second source to maintain, no drift between what humans see and what agents act on.

For your team

Understand any service, fast

A visual, navigable model you can actually trust — and edit. Onboard, review, and reason about the system without spelunking the code.

Capability map in plain English
C4 context, components, ERD & process
Drift & PR architecture review
Onboard to any service in minutes

For your agents

Grounded context over MCP

Your coding agent asks the model instead of reading files — getting the precise architectural slice it needs, with citations it can verify.

Query the model over MCP
Exact slice, ~98% fewer tokens
Grounded answers with file:line
Write back: curate & build specs

What you get

Everything that makes architecture knowable.

Capability map

What the system does, in plain English. PMs, architects, and new hires get the picture without reading code.

Trustworthy by construction

Every element is traced to file:line with a confidence score. Verifiable architecture — not plausible-looking AI guesswork.

Diagrams that stay true

C4 context, components, sequences, an ERD, and the business process — all projected from one model, so they never drift.

Drift & PR review

Your committed model vs. reality, and how a pull request reshapes the architecture — a code review at the architecture level.

MCP for agents

Your AI agent queries the same model and gets the exact slice it needs — about 98% fewer tokens than reading files.

Edit-then-build

Refine the model, emit a build spec, hand it to an agent that implements and verifies. The model is a source, not a read-out.

Grounded, or it doesn't ship

Every element shows where it came from.

Each component, capability, endpoint, and relation carries its file:line provenance and a confidence band. Low-confidence items are flagged for review, not hidden — so you can tell what's solid from what needs a look.

545

grounded elements

552

code references

86%

mean confidence

How it works

From repo to shared model in three steps.

Analyze

archmantic analyze

Reverse-engineer a grounded model from any repo — no config.

Connect

archmantic push

Generate a CLI token here, then push the model to your org.

Share & build

one model

Your team and agents read the same model; edit it and emit a build spec.

Give your team and your agents one source of truth.

Free to start. Point it at a repo and see your architecture in minutes.

Living architecture · humans + agents

The architecture model your team and your AI agents actually trust.

View on GitHub

$ npx archmantic analyze && archmantic push

archmantic · payments-api

Capability map

Charge a card● high

Refund payment● high

Dispatch webhook● medium

Reconcile ledger● medium

Context

Polyglot — TS/JS & PHP/Laravel · monorepos · APIs & data models

TypeScript

JavaScript

PHP

Next.js

React

Vue

NestJS

Express

Laravel

Inertia

Livewire

Prisma

Drizzle

GraphQL

PostgreSQL

MySQL

SQLite

One model, two audiences

The same grounded model — read as docs by people, queried as context by agents.

No second source to maintain, no drift between what humans see and what agents act on.

For your team

Understand any service, fast

A visual, navigable model you can actually trust — and edit. Onboard, review, and reason about the system without spelunking the code.

Capability map in plain English
C4 context, components, ERD & process
Drift & PR architecture review
Onboard to any service in minutes

For your agents

Grounded context over MCP

Your coding agent asks the model instead of reading files — getting the precise architectural slice it needs, with citations it can verify.

Query the model over MCP
Exact slice, ~98% fewer tokens
Grounded answers with file:line
Write back: curate & build specs

What you get

Everything that makes architecture knowable.

Capability map

What the system does, in plain English. PMs, architects, and new hires get the picture without reading code.

Trustworthy by construction

Every element is traced to file:line with a confidence score. Verifiable architecture — not plausible-looking AI guesswork.

Diagrams that stay true

C4 context, components, sequences, an ERD, and the business process — all projected from one model, so they never drift.

Drift & PR review

Your committed model vs. reality, and how a pull request reshapes the architecture — a code review at the architecture level.

MCP for agents

Your AI agent queries the same model and gets the exact slice it needs — about 98% fewer tokens than reading files.

Edit-then-build

Refine the model, emit a build spec, hand it to an agent that implements and verifies. The model is a source, not a read-out.

Grounded, or it doesn't ship

Every element shows where it came from.

545

grounded elements

552

code references

86%

mean confidence

How it works

From repo to shared model in three steps.

Analyze

archmantic analyze

Reverse-engineer a grounded model from any repo — no config.

Connect

archmantic push

Generate a CLI token here, then push the model to your org.

Share & build

one model

Your team and agents read the same model; edit it and emit a build spec.

Give your team and your agents one source of truth.

Free to start. Point it at a repo and see your architecture in minutes.