CodeTrace-AI

Ground your AI coding agent in your actual codebase

12 followers

Ground your AI coding agent in your actual codebase

12 followers

CodeTrace-AI is an autonomous engineering intelligence platform that understands entire repositories through structural analysis, semantic search, dependency mapping, call graphs, and AI reasoning.

Free

Launch tags:Open Source•Developer Tools•Artificial Intelligence

Launch Team

Wispr Flow: Dictation That Works EverywhereStop typing. Start speaking. 4x faster.

Promoted

Maker

📌

Hey — I'm Viraj, solo dev on this. CodeTrace-AI started from a frustration I kept hitting with AI coding agents on real codebases: they'd confidently refactor something and miss that a function was called via a dynamic import three files away, or invent an API that didn't exist. The agents were smart. They just weren't grounded. So I built the grounding layer. CodeTrace-AI indexes your repo locally using Tree-sitter for parsing, ChromaDB for semantic search, and a SQLite/NetworkX graph for structural relationships. It exposes all of this to your agent through an MCP server, so Claude Code / Cursor / any MCP client can actually ask "who calls this function" or "what's the blast radius of changing this signature" — and get a grounded answer instead of a plausible-sounding guess. Two things I care about that most competitors don't do: 1. **Local-first, air-gap deployable.** Your code never leaves your machine. Works fully offline. Regulated environments can actually use it. 2. **Honest about uncertainty.** Dynamic imports and config-driven wiring are hard. Instead of pretending, CodeTrace-AI labels every edge in the graph as STATIC_RESOLVED, STATIC_INFERRED, DYNAMIC_UNRESOLVED, or CONFIG_DECLARED — so your agent (and you) know how much to trust each connection. There's also an interactive HTML architecture visualizer that I honestly should have marketed harder from day one. The core is open source. A runtime-static correlation layer is coming as a paid tier. Would love feedback — especially on the MCP server integration story.

Report

2mo ago

A dependency map for test coverage would be a great addition, showing which unit and integration tests actually exercise each module or function call chain. Right now I can see what calls what, but not what gets validated, which makes it harder to spot untested critical paths before refactoring. Could your AI flag functions with no direct or indirect test coverage and suggest missing test cases for high-risk code?

Report

2mo ago

Maker

@nuriyebavatll2 Sorry for late reply.

Thanks so much for this feedback! You hit the nail on the head regarding one of the biggest anxieties developers face with AI-assisted refactoring: "I know what calls this, but will a test catch it if something breaks?"

Mapping unit and integration tests directly onto the dependency graph is a natural extension of our blast-radius engine. We’re actively exploring an overlay where CodeTrace connects test suites to execution paths, highlighting "blind spots" (critical, high-blast-radius functions with zero direct/indirect test coverage) and prompting the AI agent to draft targeted tests before allowing refactors.

This is definitely going straight to the top of our roadmap. Would love to have you in our Slack, GitHub if you’d like early access to test this when we drop the beta! 🚀

Report

16d ago

would love to see a way to drop in an issue or bug report and have it auto-trace the failing call path back to recent commits that touched those files, basically a blame + impact overlay. that would save so much time compared to hunting through git history manually

Report

2mo ago

Maker

@melihacelercmv Sorry for late reply.

This is a brilliant idea! Triage and root-cause analysis are huge time sinks, especially when a bug touches multiple layers across several recent commits.

Having an Issue/Stack Trace → Call Path → Git Blame + Blast Radius pipeline fits right into CodeTrace's architecture. Since we already have the local graph of your codebase, feeding in an error log or issue and correlating the execution path with recent commit deltas (git log/blame) to pinpoint the exact suspect commit is completely doable — and 100% local!

Adding this "Time-Travel Debugging / Blame Overlay" to our roadmap. Thanks for the awesome suggestion, and feel free to jump into our GitHub/community if you'd like to share sample workflows you'd want supported! 🔥

GitHub, Slack

Report

16d ago

Reviews