Agent-Up
Agent-Up is a local runtime control plane for parallel AI-assisted software development.
It manages the running development environment around applications: workspaces, Git worktrees, application processes, port allocation, Docker services, isolated browser profiles, logs, diagnostics, event history, and automation surfaces.
Agent-Up is not an application framework, deployment tool, IDE, Docker replacement, Git replacement, or production orchestrator.
Development Preview: Agent-Up is under active development. CI produces preliminary platform artifacts, but APIs, configuration, installer behavior, and workflows may change without notice.
Every AI agent can work in its own Git worktree with its own branch, runtime environment, browser profile, infrastructure, application state, diagnostics, and event history. Agent-Up is being built to make switching between these isolated workspaces practical while letting developers and AI agents interact with the same running applications.
What Agent-Up Solves
Modern AI-assisted development creates multiple parallel runtimes. Existing tooling usually assumes one developer running one application instance, which leads to:
- Constantly starting and stopping services.
- Docker infrastructure collisions.
- Browser tab sprawl.
- Duplicated authentication flows.
- Inconsistent runtime state.
- Manual process management.
- Difficult validation of AI-generated changes.
Agent-Up solves these problems without requiring changes to application source code.
System Shape
The Server is the single source of truth. Desktop, CLI, and MCP clients are all thin clients over server-owned runtime state.
+----------------------+
| AgentUp.Server |
|----------------------|
| Workspace Manager |
| Process Manager |
| Browser Manager |
| Port Manager |
| Event Recorder |
| Diagnostics |
| Playwright Generator |
| MCP Server |
+----------+-----------+
|
+------------------+-------------------+
| | |
+---------------+ +---------------+ +------------------+
| Avalonia UI | | AgentUp CLI | | MCP Clients |
| Human UI | | Thin Wrapper | | ChatGPT |
| | | | | Claude / Codex |
+---------------+ +---------------+ +------------------+
Getting Started
See Current Limitations for the current implementation status of each major area.
1. Start the server
Installed Desktop artifacts run the Server as the local agent-up-server service.
From the repository root:
dotnet run --project AgentUp.Server
The server starts on http://localhost:5000 in the current development launch profile.
2. Add an agent-up.json to your project
Create agent-up.json in the root of the repository you want to manage:
{
"name": "My App",
"applications": [
{
"name": "Frontend",
"command": "npm run dev",
"path": "."
}
]
}
3. Push the workspace definition
From your project directory, run the CLI with dotnet run:
dotnet run --project /path/to/AgentUp.CLI -- start --server http://localhost:5000
Or set the server URL once as an environment variable and omit --server on every call:
export AGENTUP_SERVER_URL=http://localhost:5000
dotnet run --project /path/to/AgentUp.CLI -- start
This reads agent-up.json, captures the current git branch and commit, and pushes the workspace and application definitions to the server. The server then exposes them at GET /api/workspaces/{id}/applications for the desktop app to consume.
Documentation Map
- Workspace defines the workspace model.
- Setup describes the current source-first development workflow.
- Releases describes packaged artifacts and MinIO/S3 release upload.
- Current Limitations explains what is implemented, experimental, in progress, and planned.
- Browser covers shared browser sessions and automation.
- Browser Profiles explains per-workspace browser isolation.
- CLI covers the human-friendly command wrapper.
- Configuration and agent-up.json describe declarative application setup.
- Roadmap captures the long-term direction.
Implementation details live in the Developer Guide.