Stop fixing what AI breaks.

Aspect Code analyzes your codebase and generates an informed AGENTS.md — then watches for changes and keeps it up to date and optimized.

~/my-project — aspectcode

   __ _ ___ _ __  ___ __| |_  __ ___  __| |___
  / _` (_-<| '_ \/ -_) _|  _|/ _/ _ \/ _` / -_)
  \__,_/__/| .__/\___\__|\__|\__\___/\__,_\___|
           |_|
  config loaded · API key found · evaluator on
  * Watching  123 files · 200 edges · openai · 15.0s
  3/8 probes passed · 4 fixes applied
  6.5k in · 762 out
  ├ 5 sections · 22 rules · 6 file-specific guidelines
  └ covers: backend/app/core/config.py, backend/app/models.py, +3 more

Benchmarked:50%fewer failures·~27%smaller diffsDetails

Get started Quick-start guide

Open-source · Offline-first · No API key required

Output

One file your AI assistant reads automatically

Generated offline from static analysis — no API key required. Add an API key to unlock LLM-powered evaluation and iterative improvements.

Files added to your repo

AGENTS.md     ← always generated (no API key needed)

# Optional (--kb flag):
kb.md         ← detailed knowledge base with full analysis

AGENTS.md excerpt

# AGENTS.md — generated by Aspect Code

## Golden Rules
- Never edit lib/constants.ts without checking all 38 dependents
- All API routes go in app/api/ — never in lib/ or utils/
- Skip migration files → migration jobs handle schema changes

## Project Architecture
FastAPI backend + React frontend monorepo. 123 files, 200 edges.
Entry points: backend/app/main.py, frontend/src/main.tsx

## High-Risk Hubs
- lib/constants.ts (38 dependents)
- models/User.ts (22 dependents)
- middleware/auth.ts (15 dependents)

## Coding Conventions
- snake_case for Python, camelCase for TypeScript
- All new endpoints need tests in tests/api/

## When Changing Code
- models/User.ts ↔ api/users.ts ↔ tests/api/users.test.ts
- lib/auth.ts ↔ middleware/auth.ts ↔ config/auth.ts

Reviewable markdown

Commit or gitignore

Auto-updates on save (watch mode)

No API key required

Benchmarks

Measured improvement

Tested on FastAPI and Django repos with Opus 4.5. Plain prompts vs. Aspect Code context.

50%

fewer failures

−24 → +4

net tests

~27%

smaller diffs

0 → 3

clarifications

Read the benchmark writeup

Results vary by repo and model. Related research

Ready to try it?

No API key required — install and generate your first AGENTS.md in under a minute. Add an API key later to unlock LLM-powered evaluation.

$ npm i -g aspectcode && aspectcode

Get started Quick-start guide