llmcc: a very fast tool brings multi-depth architecture views for coding agents

[allenanswerzq]

Problem

Understanding large codebases is hard. Current approaches have tradeoffs:

• LSP/grep/ast-grep — Great for local search, but requires many rounds of back-and-forth to build a mental map
• RAG/vector-search — Can be slow, expensive to build, and lags behind code changes

Its very nature for coding agents (human also) to grasp architecture quickly before diving into details. There's a gap for fast, token-efficient structural understanding.

---

llmcc

What it does: Generates multi-depth architectural view of a codebase in extremely faster speed.

Benefits:

• Fast — Full codex-rs analysis in ~0.5s, millions level can be done in seconds.
• Better understanding — Multiple depths graphs to provide very comprehensive context insights
• Token-efficient — Save many back and forth grep steps

Repo: https://github.com/allenanswerzq/llmcc

---

Example: codex-rs

Samples: llmcc/sample/rust/codex-pagerank
Full graph: depth_3_file.svg (download and open in browser for human to see, model feeds .dot directly)

Question (one shot/seconds time): ask about the codebase architecture

Question (one shot/seconds time): "Where should I look to modify prompt handling?"

---

Would love feedback from the codex team, if this is something worth spend more time on, thanks!