Skip to main content

Agent-Up

Agent-Up is a local runtime control plane for parallel AI-assisted software development. It is not an application framework, deployment tool, IDE, or production orchestrator.

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.

What Agent-Up solves​

Modern AI-assisted development creates multiple parallel runtimes. Existing tooling usually assumes one developer running one application instance.

  1. Stop colliding runtimes

    Constantly starting and stopping services, Docker collisions, and browser tab sprawl.

  2. Keep one source of truth

    The Server owns orchestration. Desktop, Mobile, CLI, and MCP clients stay thin.

  3. Leave application source unchanged

    Agent-Up injects ports and environment at launch. Applications do not take an Agent-Up dependency.

Clients

Desktop

Human workspace chrome

Mobile

Android, iOS, and the installable PWA

CLI

agent-up

MCP

Five named servers, never a shared /mcp

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 listens on http://localhost:5001.

  2. Add agent-up.json

    Put it at the root of the repository you want to manage. Only name is required.

  3. Start the workspace

    From that repository, run agent-up start --server http://localhost:5001.

{
"name": "My App",
"applications": [
{
"name": "Frontend",
"command": "npm run dev",
"path": ".",
"ports": [
{ "variable": "PORT", "defaultPort": 3000 }
]
}
]
}

When invoking with dotnet run, pass CLI arguments after --:

dotnet run --project /path/to/AgentUp.CLI -- start --server http://localhost:5001

Documentation map​

Docs are grouped by product slice, not by client.

  • Start covers downloads, setup, releases, limitations, and the roadmap.
  • Workspaces is identity, connect, start/stop, and clones.
  • Applications is processes, ports, console, and hosted GUI apps.
  • Git is working-tree review, history, and remotes.
  • Commits is the agent proposal queue.
  • Agents is the ACP session and subscription sign-in.
  • Browser is isolated surfaces and validation.
  • Diagnostics is console, health, and audit history.
  • Verification is path-rule checks and receipts.
  • Configuration is agent-up.json.

Implementation details live in the Developer Guide.