proxysoul/Empryo

What it solves

Empryo addresses the inefficiency of traditional AI coding agents that rely on simple text searches (grep) or reading entire files, which often leads to a lack of understanding of code dependencies and high token costs. It solves the problem of "blast radius"—knowing exactly what will break when a specific piece of code is changed—before the edit occurs.

How it works

Empryo uses a graph-powered approach to codebase understanding. Upon launch, it uses tree-sitter to parse the repository into a live dependency graph of symbols, imports, and call sites, ranked by PageRank and git co-change data. This allows the agent to navigate the codebase via graph queries rather than burning LLM tokens on reading files. It performs structural edits directly through the Abstract Syntax Tree (AST) using over 65 symbol-level operations, ensuring changes are atomic and independent of whitespace.

Who it’s for

It is designed for developers who want an AI coding assistant that understands the structural context of their entire project, supports multiple programming languages (30+), and offers flexibility in choosing which LLM models to use for different tasks (scouting, writing, reviewing).

Highlights

• Code Genome: A live dependency graph that enables millisecond searches and blast-radius analysis without LLM token spend.
• Multi-Agent Routing: A system where different roles (e.g., scout, coder, reviewer) can be assigned to different models from 22+ providers.
• AST-Based Editing: Structural edits across 30+ languages with atomic batches and all-or-nothing rollback.
• Time Machine: Integrates prompts with git checkpoints to allow rewinding both code and conversation together.
• Privacy-First: Runs locally on the user's machine with "bring your own key" or fully local model support via Ollama/LM Studio.