Introduction

Rusty is a desktop IDE for working with coding agents. Instead of a single chat window, it gives agents a structure: workflows that break a job into steps (research, design, build, verify), a task graph you can see, permission gates on risky actions, and diffs you review before you commit.

Under the hood, Rusty runs a local agent engine that talks to the model providers you connect, so you keep your own keys, your own models and your own files.

Note

Rusty is a Developer Preview. Rusty Canvas is marked beta, and risk review and JEV decisions are marked experimental in the app. Expect things to change between releases.

The core ideas

Workspace. A folder you open in Rusty, usually a Git repository. Everything Rusty keeps for that project (chats, canvases, workflows, profiles, skills) lives in a .rusty/ folder inside it. See The .rusty folder.

Agent Mode. A chat with an agent that can read your code, edit files, run commands and search the web, subject to the permissions you set. A chat runs as a single agent by default, or follows a workflow you pick. See Agent Mode.

Workflows. A workflow decides which steps run and in what order, what each step may use, and what must be checked. Rusty ships 15: short stages that stop at a result you can read and steer, and pipelines that run a whole job. You can pick one, or let Rusty choose for each message and hand a run over to another when what it finds changes the plan. See Workflows, Built-in workflows and Auto flows.

Profiles. A profile decides what one workflow step may do: its instructions, which tools it can use, its limits, and what it must check before it can finish. Workflows and profiles are plain JSON, edited in the Behaviors tab. See Profiles & workflows.

Rusty Canvas (beta). A visual graph for larger features. A Global Explorer proposes tasks, each task gets only the context you connect to it, and results land in an isolated virtual workspace for review. See Rusty Canvas.

Skills. Reusable definitions of how a model should work: system instructions, a preferred model, and tool access. See Skills.

Providers. Rusty includes no model. You connect OpenAI, Anthropic, OpenRouter, GitHub Copilot, OpenAI Codex, or any OpenAI-compatible endpoint. See Model providers.

A tour of the interface

The rail on the left edge is your map. From top to bottom:

Icon	What it opens
Open Workspace	Pick a folder to work in, or reopen a recent one.
Files	The file explorer.
Source Control	Staged and unstaged changes, commits, branches and history. See Source control.
Rusty Canvas (Beta)	The task graph. See Rusty Canvas.
Agent Mode	Chat with an agent. See Agent Mode.
LLM Integrations	The LLM Connection Center, where you connect model providers.
Skills	Create and edit skills.
Behaviors	Design workflows and profiles.
MCP Integration	Connect MCP servers. See MCP servers.
Rusty Guide	The in-app guide to automatic model selection, smart tools, workflows and Canvas.
Token Metrics	Daily token usage. See Observability & metrics.
Settings	Appearance, editor, intelligence and system settings.

Files open as tabs in the main area next to Agent Mode, Canvas and the other views. Select code in a file and press ⌘ Enter to ask about it in place with Inline Chat.

Where to go next

1. Install Rusty.
2. Follow the Quickstart to connect a model and run your first task.
3. Read Workflows to run a job as steps you can follow, and Profiles & workflows when you want to change how agents behave.