What is Agent Code?

Agent Code is an open-source, AI-powered coding agent for the terminal. You describe what you want; it reads your codebase, runs commands, edits files, and iterates until the task is done.

It is written in Rust (single static binary), MIT licensed, multi-provider, and does not phone home unless you opt into telemetry.

What can it do?

Search, read, trace dependencies, and explain how systems work. Create files, targeted edits, apply_patch, refactors, and multi-file changes. Shell, tests, git, builds — with permissions and optional OS sandbox. Tool loops, subagents, plan mode, skills, MCP, and session resume.

How it works

You type a request (TUI, -p, ACP, or HTTP)
The query engine sends context + history to your configured LLM
The model streams text and tool calls
Tools run under the permission system (and optional sandbox)
Results feed back into the model until the turn finishes

You: "add input validation to the signup endpoint"

Agent:
  → FileRead src/routes/signup.rs
  → Grep "validate" src/
  → FileEdit src/routes/signup.rs
  → Bash "cargo test"
  ✓ Tests pass. Validation added.

Surfaces

Surface	Command	Role
Interactive TUI	`agent`	Fullscreen transcript, modals, queue, tasks
Headless	`agent -p "…"`	Scripts and CI
HTTP	`agent --serve`	Local API / Flutter client
ACP	`agent acp`	IDE stdio bridge
Security scan	`agent security-scan`	Whole-repo AMR scan

Key features

Modern fullscreen TUI with streaming, HITL modals, mode badge, prompt queue
30+ tools — files, shell, search, worktrees, web, LSP, MCP, subagents, monitors, cron
Multi-provider — Anthropic, OpenAI, Azure, xAI, Google, OpenRouter, local OpenAI-compatible, and more
Skills & plugins — reusable workflows and project conventions (AGENTS.md)
Permissions — ask / allow / plan / accept_edits / auto, protected dirs, bypass lock
Sessions — save, resume, fork, rewind, compact
Security-scan — plan → shard → map → reduce vulnerability hunting
Privacy default — no product telemetry unless you opt in

Where to go next