Skip to content
AG.IDE AG.IDE Docs

AG.IDE

An IDE for agents

Index. Rank. Understand. Improve. Repeat.

AG.IDE helps coding agents holistically understand a codebase from a code quality perspective. It indexes the whole repository, ranks the worst offenders, shows the graph and history around each target, and gives agents the context they need to improve the system iteratively.

The Agent Loop

Section titled “The Agent Loop”
  1. 1Index the whole repository.
    Build a local map of files, symbols, calls, history, and quality signals.
  2. 2Rank the worst offenders.
    Start with the functions most likely to pay down debt: complexity, coupling, churn, side effects, and ambiguity.
  3. 3Understand the blast radius.
    Follow callers, callees, file anatomy, impact, and history before asking an agent to edit.
  4. 4Improve with focused context.
    Give the agent one target plus the surrounding graph instead of the whole codebase.
  5. 5Review the patch.
    Map the diff back to indexed symbols and risk signals before accepting the change.
  6. 6Re-index and repeat.
    Score the repository again, find the next best target, and keep quality climbing.

Project Overview

Section titled “Project Overview”
AG.IDE project overview for Redux with history chart and ranked functions
The project overview combines history, threshold trends, file navigation, and the ranked function table.

Quick start

Run the agent loop: index, rank, inspect the graph, review a patch, and repeat.

CLI reference

Use AG.IDE from terminals, scripts, and agent workflows with readable shell output.

Desktop UI

See the ranked overview, quality history, file tree, code heatmap, and symbol detail views.

What AG.IDE Gives Agents

Section titled “What AG.IDE Gives Agents”

Whole-codebase memory

Files, packages, symbols, call edges, history, and quality metrics in a local index.

Debt ranking

A prioritized list of targets where refactoring is most likely to improve the system.

Graph context

Callers, callees, impact, and file anatomy so an agent knows what it is touching.

Focused prompts

The right target with the right surrounding context, instead of a blind request to edit code.

Patch review

Diffs tied back to indexed symbols and risk signals before the change enters the codebase.

Second loop

A continuous index-rank-refactor cadence beside the normal write-review-merge workflow.