Archie · architecture analysis for AI agents

Agent-built codebases
tend to erode over time.

Without an architecturally sound foundation, they can quickly become unmaintainable, and the speed agents bring to development starts to fade. That's why we built Archie: live semantic documentation for your codebase that also acts as an enforcement layer.

You can't build what you can't understand.

Analyze your first repo →

01. THE THESIS

Archie builds semantic understanding of your codebase.

Curated knowledge, delivered exactly where and when agents need it.

Without Archie

Agents grep to guess

Every session starts with the agent searching your codebase to reverse-engineer the rules.

With Archie

Curated answers, ready

The right context surfaces at the file the agent is touching. No grep, no guessing, no drift.

The payoff

Agents inherit how it was built.

Every architectural decision, every pattern, every piece of reasoning — preserved and followed automatically. Not just the code.

02. UNDER THE HOOD

Multi-wave AI analysis.
Semantic enforcement.

Archie runs the same analysis a senior architect would — then embeds the conclusions where agents work.

Build phase

/archie-deep-scan

STEP 1 — Structural index

Indexes the whole repo

File tree, import graph, framework signals, skeletons of every source file (signatures + imports). Deterministic, no AI yet — the foundation that grounds everything else.

Step 2 — Wave 1Grounded reads

Parallel specialist agents

Each agent reads the index + pulls source files on demand to ground every claim in real code.

Structure

components, layers, placement

Patterns

communication, design patterns

Technology

stack, deployment, dev rules

UI Layer

state, routing (if frontend)

STEP 3 — Wave 2

Architectural reasoning

decision chains, trade-offs, pitfalls with causal links

Step 4 — Rule generation

Rule synthesis → rules.json

every rule carries severity_class, WHY, and EXAMPLE — semantic content the agent reads at edit time

STEP 5Optional

Intent layer

per-folder CLAUDE.md generated via bottom-up DAG

One-time, ~15 min

Enforce phase

Every agent edit

> .archie/rules.jsonrule

name:

domain_layer_boundary

severity_class:

decision_violation

DESCRIPTION

Domain layer must not import from
infrastructure or API layers.

WHY

Forced by: clean architecture decision.
Enables: independent testability of
business logic, swappable adapters.

EXAMPLE

✓ from domain.entities import User

✗ from infrastructure.db import Session

Severity gates

decision_violation · pitfall_triggered · mechanical_violation

Blocks (exit 2)

tradeoff_undermined

Warns prominently

pattern_divergence

Informs quietly

/archie-scanOn-demand grounding1-3 min, run often

Three specialist agents (architecture, health, patterns) pull files on demand against the existing index to ground every finding in real code. New findings become new rules. The model sharpens with every run.

03. RECEIPTS

This is what semantic understanding looks like.

A real Archie blueprint, embedded below. Click to open the live, navigable viewer.

Open the blueprint

04. OUTCOMES

Ship faster.
Ship safer.

Better codebase → better uptime → more money in the bank. Here's how Archie compounds into dollars.

Stable release cadence

Agents land grounded code consistently. Predictable shipping, no surprise firedrills derailing the schedule.

Less downtime cost

Architectural mistakes caught at edit time, not at 3am. Real uptime, real money saved on incidents.

Cut validation overhead

Agents land grounded code. Reclaim weeks of dev time from validating AI output.

Less pointless complexity

Agents respect existing patterns instead of inventing new ones. Fewer bad decisions and bad code make it into production.

Stop watching your codebase erode.

Three minutes to install. Compounding returns from day one.

npx @bitraptors/archie .

Currently atop Claude Code · others coming