A self-hosted AI workspace control plane
TaskHandoffThe self-hosted AI development workbench
Run Codex and OpenCode in one self-hosted workbench for local and remote AI development, workspace management, and code review
Install on your desktop or deploy your own server
Built around the tools you already use
Agent skill
One line teaches your agent to run TaskHandoff
Install the skill and your agent knows how to connect to the control plane, drive sessions, and diagnose failures
Copy this into your agent chat
Install the TaskHandoff skill: https://docs.thandoff.com/.well-known/skills/index.jsonOne workbench for your whole development flow
Bring your conversations, code, and environments together. Less switching. More focus
Sessions stay clear
Compact mode focuses on the current turn; switch to the full conversation when you need to audit
Learn moreWorkspaces stay isolated
Push several tasks at once with instances and worktrees, without files or branches colliding
Learn moreProgress everywhere
Reach the same control plane from desktop, web, mobile, and chat bridges
Learn moreYour work belongs together Not in scattered windows
From tasks to sessions, from local to remote. Find every perspective in the same workbench

Instances, sessions, and controls. Together in one workbench
Manage the whole job, not just the conversation
Every step has a clear environment, execution state, and delivery path. Focus on decisions while TaskHandoff keeps the process together
- 01
Connect your environment
Add a local or remote node, then configure models and repositories
- 02
Create an isolated instance
Start from a Git repository or local folder with a Local or Docker runtime
- 03
Delegate the task
Choose the agent, model, fixed working directory, and permission mode
- 04
Review and deliver
Check tool activity and Git changes, then ship via branch or worktree
A few things worth knowing
Understand how it runs. Choose the environment that fits
Do I need Docker?
Not necessarily. Local Runtime uses tools and directories on the host. Choose Docker for multiple isolated instances. Each host user can run only one Local Runtime instance
Can I use Codex and OpenCode together?
Yes. Create and manage sessions for both agents in one workbench. Each session uses its selected agent and model; some advanced capabilities differ between agents
Will tasks continue after I close the browser?
Yes. Tasks run in controlled instances, not in the browser tab. Stopping an instance, closing a session, or stopping its runtime services can interrupt work. Quitting the desktop app is also different from closing a tab
Does compact mode remove conversation history?
No. It changes only how the session is displayed, not its history or agent context. Switch to conversation mode whenever you need the full record
Can I access remote tasks from my phone?
Yes, for running sessions and instance operations. Configure authentication and HTTPS before remote access. Use Web or Desktop for creating instances and full infrastructure management
Your next task Start it here
Connect your environment. Choose your agent. Give every task its own workspace, and keep the important decisions in your hands
Install on your desktop or deploy your own server