Desktop App
The AgentLoop desktop app is the primary way to use AgentLoop. It’s a cross-platform GUI that hosts the chat surface, kanban board, agent workbench, and the rest of the workspace in a single window — backed by a local background daemon that does the actual work.
Launching the App
Download AgentLoop and install it for your platform, then open it the same way you’d launch any other desktop application:
- macOS — Open AgentLoop from Applications or Spotlight
- Windows — Coming soon
- Linux — Coming soon
On first launch, you’ll go through sign-in and model selection. See Getting Started for the walkthrough.
The daemon that runs your agents starts automatically when you open the app. You don’t need to start it manually — see Background Daemon for how its lifecycle works.
Workspace Layout
The desktop window is organized into three regions:
- Sidebar (left) — Project list, chat threads, and the navigation rail
- Workspace (center / right) — Chat (the home surface) or the active view
- Title bar (top) — Account menu, update notifications, and global controls
The workspace defaults to Chat. Pick a view from the navigation rail to swap the main pane, or pop chat out into a floating overlay. A collapsible chat dock sits at the bottom of every view so you can keep talking to AgentLoop without leaving the view you’re in.
The Chat Surface
Chat is how you tell AgentLoop what you want. Type a message and the chat agent figures out what to do — usually delegating to the Product Manager to break the work into tasks, or pulling in another agent for a specific request.
Examples of things you can ask:
- “Break down user authentication into tasks and add them to the board.”
- “What’s the status of task 5?”
- “Have the engineer implement the login API endpoint.”
- “Ask QA to test the new authentication flow.”
- “Add a dependency: task 3 must finish before task 7 can start.”
- “Show me what’s blocked and why.”
Chat threads are persisted, so you can pick up where you left off after closing the app.
Views
You switch views from the navigation rail in the sidebar. Chat is the home surface; everything else is one click away. Each view has its own guide — follow the links below.
Primary views
| View | What it shows |
|---|---|
| Chat | The conversational home surface. Talk to AgentLoop in plain language, watch tool calls stream live, and answer inline prompts (like spec questionnaires) without leaving the conversation. |
| Tasks | The kanban task board for the active project. Switch between Board, List, Table, and Graph (dependency DAG) modes, drag cards between lanes, and open a task’s detail dialog. |
| Workflows | A visual node-graph editor for composing cross-agent automations — triggers (cron, webhook, polling, kanban-column entry, manual run), routing, and actions on one canvas. Run and validate them from the Executions and Tests tabs. |
| Agents | The agent workbench. Browse the roster of agent types, then drill into any agent for its Overview, Behavior Tree, Tools, MCP, Instances, and Diagnostics. Opening a running instance shows its live behavior-tree visualization alongside its transcript, logs, and blackboard. |
| Specs | The spec library — a searchable collection of your project specs with a markdown viewer, revision history, import, and diff. |
Secondary views
| View | What it shows |
|---|---|
| Alerts | System notifications and workflow alerts, with an unread badge on the rail. |
| Skills | Composable, markdown-defined skills you can enable or disable per agent. |
| Config | Runtime settings, guardrails, and model selection for the active project. |
The standalone Behavior Tree and Execution DAG panels have been folded in: behavior trees now live inside each agent in the Agents view, and the dependency graph is the Graph mode of the Tasks view.
Projects and Subprojects
The sidebar’s Project List lets you manage multiple projects in one window. AgentLoop discovers your projects on disk and remembers which ones you’ve opened.
Within a project you can also create subprojects — scoped slices of work (for example, a single feature or sprint) with their own kanban lane chips and runner controls. Subprojects are useful when you want a few separate work-streams visible side by side without juggling multiple repos.
Running Agents
AgentLoop has per-panel runner toggles that control whether agents are actively picking up work for a project or subproject. When the toggle is green (running), the orchestrator is scheduling tasks; when it’s grey (paused), no new agent work starts but anything in flight finishes.
This gives you fine-grained control: you can leave one project’s runner on overnight while keeping another paused.
System Tray
AgentLoop runs from the system tray. Closing the main window doesn’t quit the app — the daemon and any in-progress agents keep running, and you can re-open the window from the tray icon.
To fully quit AgentLoop, use Quit from the tray menu (or the standard app menu).
Auto-Update
AgentLoop checks for updates in the background. When a new release is available, you’ll see an update banner in the title bar; you can install it on your next restart, or defer until later.
Spec Workflow (Project Intake)
When you open a new project, AgentLoop can run a guided spec workflow — an interactive questionnaire that helps you describe what you want to build before any tasks are created. It’s optional; you can skip it and just start chatting.
Account & Models
The account menu in the title bar handles authentication and model selection. You can:
- Sign in with your AgentLoop account
- Switch between AgentLoop subscription credits, free tier models, and your own provider API keys
- Connect Anthropic, OpenAI, Google, OpenRouter, and other providers
See Getting Started for the full setup walkthrough.