Egent Code Plexus

What egent-code-plexus does

egent-code-plexus builds a structural graph from a code repository for autonomous coding workflows. It records symbols and relationships such as calls and inheritance, then exposes focused queries for locating code, inspecting nearby graph context, tracing downstream effects, and examining application routes.

The output is designed for agent context windows. Results use TOON or compact JSON and focus on symbols, source lines, and graph edges instead of prose or presentation-oriented formatting. When static analysis cannot resolve a relationship, the graph emits a typed BlindSpot record rather than presenting an inferred edge as fact.

The parser covers 31 languages, including application code, infrastructure definitions, SQL, Dockerfiles, GitHub Actions, Terraform, Move, and Solidity. The project is aimed at the Coding Agents category and is most relevant when an agent needs structural repository knowledge during implementation or refactoring.

How it works

Indexing parses the repository, resolves relationships, and serializes the result into a graph file. Each query then memory-maps that file and exits after returning its result. This stateless, one-shot model does not require a long-running daemon or a graph warm-up process.

The repository describes measured query times below 175 ms, including process startup, on a 22,645-file polyglot corpus. Its reported cold ingest for that corpus is 2.60 seconds. These figures come from specified benchmark hardware and test repositories, so they should be treated as reference measurements rather than a guarantee for every codebase.

Setup and configuration

Linux and macOS installation is available through the project’s release installer:

curl -sSfL https://github.com/coseto6125/egent-code-plexus/releases/latest/download/install.sh | sh

The project is implemented in Rust and is distributed under the MIT license. The repository also provides benchmark scripts for reproducing indexing and query measurements.

Tools and capabilities

The documented query surface includes:

• summary for repository registry information, with a detailed mode for additional framework confidence scoring.
• find for symbol and text-oriented search, including a BM25 mode.
• inspect for symbol resolution and one-hop graph traversal.
• impact for breadth-first caller or dependency analysis, with upward and downward directions and a Git baseline mode.
• routes for declarative and imperative route enumeration.
• cypher for pattern-based graph queries.

Impact analysis follows recorded Calls and Extends edges. The Git baseline mode combines changed-file detection with parsing and graph traversal, making it suitable for estimating the reach of a change.

Limitations and notes

The project describes itself as an early release. Its graph is based on static analysis, so dynamic dispatch, reflection, and unresolved imports can remain unknown; these cases appear as BlindSpot records. The tool is not presented as a drop-in replacement for GitNexus: egent-code-plexus targets autonomous agents with a stateless CLI, while GitNexus targets human developers and IDE integration.

Performance and language coverage depend on the repository and source constructs being analyzed. Benchmark comparisons also use capped file sizes and particular test machines, repositories, and tool versions.