LLMs can't fit a complex codebase in context — so they guess file by file, on outdated versions. CDX maps your entire repository and the exact dependency versions you run — eliminating hallucination with end-to-end visibility through call trees across dependencies.
Part of the AIvantGuard SDK — tier-based licensing, usable standalone or as an MCP service.
Agents search private repositories file by file, losing the structure that connects files. The big picture is never in view.
Models are trained on a snapshot of a library — but your project runs a different version. Interfaces drift; models hallucinate and burn time in trial-and-error.
A map of the entire codebase with the exact versions of every dependency — AI works with complete structure and the interfaces you actually have. Heterogeneous platforms? CDX fuses their code into the best solution.
One map of the whole codebase — not file-by-file search.
456 experts in an MoE architecture, each trained on a specific crate version.
Negative-weight training blocks known vulnerability patterns.
Scan for vulnerability patterns before you commit.
Same behavior in different languages, detected automatically.
Each dependency matched to the expert trained on your version.
12+ tools, stdio JSON-RPC — works with Claude Code, OpenCode, Codex.
Composes the best solution from what each platform does well.
Zero Python in production. Candle + tree-sitter. Cross-language concept search across 6 languages.
Trained experts
MCP tools
Languages
Rust — zero Python in production
| Version | Milestone | Meaning |
|---|---|---|
| v0.1 | Foundation | Pipeline, 8 languages, model scaling |
| v0.3 | Agent ready | MCP server — agents call it as a tool |
| v3.0 | Security | CVE-aware generation in version experts |
| v5.0 | Rust native | Zero Python in production |
| v6.0 | Cross-language | Tauri IPC trace, concept search, 6 languages |
Now
CLI and MCP server running on real repositories in daily use. Core engine, experts and security pipeline live.
Next
Wrapping engine, experts and MCP into a distributable product — installers, packaging, release tooling.
Then
Integrations with agent platforms, CI pipelines and IDEs — wherever code is written.
Long term
General availability. Today the engine is under testing — not yet on sale.
Repository context through MCP — generation that matches the codebase.
Vulnerability-pattern scanning before commit, CVE-aware generation.
Generation, refactoring candidates, duplication detection across languages.
Behavior-true cross-language migration — confidence, not guesswork.