Aspect Code is live. One command, every repo.
Aspect Code generates and maintains AI coding context automatically. Inspired by autoresearch, built for every agent.
Aspect Code is now publicly available.
npm i -g aspectcode && aspectcodeThat's it. One command. It analyzes your codebase with tree-sitter, generates AGENTS.md and scoped rules, then watches your changes and gets smarter in the background. Works with Claude Code, Cursor, Copilot, Codex, Windsurf, Cline, Gemini, and Aider.
Why now
Karpathy's autoresearch has been out for a while now. The core insight: LLM agents work dramatically better when they have structured context about what they're working on. Build a knowledge scaffold before the agent starts writing. Agents don't fail because they're dumb. They fail because they're blind. Aspect Code does the same thing for codebases.
Today, Anthropic's internal KAIROS system leaked. It's Claude Code's context engine — the thing that makes Claude Code good on repos it's never seen. It builds a dependency graph, identifies high-risk files, and generates scoped instructions.
The big labs are building this internally. Aspect Code makes it available to everyone, for every agent, right now.
What makes this different from a markdown generator
Repomix, SummonAI Kit, GitNexus — they all generate a context file. Run once, done. That's step 1.
Aspect Code does step 1, then keeps going:
- Watch mode monitors every file change against the dependency graph
- Auto-resolve uses LLM judgment to handle routine warnings without interrupting you
- Preferences learn from your corrections — the same warning never fires twice
- Dream cycles consolidate your feedback into better AGENTS.md autonomously
- Probe-and-refine tests the generated context against your actual codebase (40% fewer errors on SWE-bench)
The context doesn't just get generated. It gets maintained and improved. Set it up once. It handles the rest.
Try it
npm i -g aspectcode
aspectcode login
aspectcodeFree tier: 100K lifetime tokens. No credit card.