Summary
I would like a way to run OpenWork like a private cloud on my own server.
My main setup is this: I have a server that runs my models and workspaces. It might use Ollama, vLLM, LM Studio, LiteLLM, or another OpenAI compatible endpoint. I want to connect to that server from the OpenWork desktop app on another computer, start agent work, close my laptop, and come back later to see the same sessions, logs, diffs, approvals, and progress.
Problem / goal
This is not just remote model access. I want the actual agent work to keep running on the server.
For example, I connect OpenWork to my private worker server. The server stores the sessions, worktrees, logs, automations, MCP servers, and approvals. A local model does most of the work because it is fast and cheap. A stronger model can plan, review, or check the local model’s work. If I close my laptop, the server keeps going. When I reconnect, I can pick up where I left off.
A lot of people in local model communities talk about this. Local models are getting good, but they need a better harness around them. They need structure, permissions, review, and good context handling.
Primary user(s)
OpenCode primitive alignment
This seems like it could build on existing pieces: OpenCode sessions, permission requests, providers and models, MCP servers, skills, plugins, and remote workers.
The new part would be an OpenWork workflow layer for model roles, such as executor model, planner model, reviewer model, fallback model, small local model for normal work, and larger model for checking important work.
Some useful workflows would be: local model does the work and a stronger model reviews it, stronger model makes the plan and local model executes it, local model works until something fails and then escalates, or reviewer model checks diffs before the task is marked done. Human approval should still apply where configured.
Alignment with VISION/PRINCIPLES/PRODUCT
This feels aligned with OpenWork being local first, cloud ready, and able to connect to remote workers.
It would let people run their own private OpenWork setup without needing hosted cloud workers. The server owns the work, and the desktop app becomes the control panel.
Testability
Run an OpenWork worker on a server.
Connect from the desktop app.
Add a local model provider like Ollama or vLLM.
Start an agent session.
Close the desktop app.
Reopen it and reconnect.
Confirm the session, logs, diffs, approvals, and progress are still there.
Then test a workflow where a local model makes a change and a reviewer model checks it.
Ready to build it yourself?
Yes
Additional context
The main idea is that the server should own the work, and the desktop app should just be the control panel. That way I can use my own hardware, keep things private, and still have a good GUI for agent work.
Summary
I would like a way to run OpenWork like a private cloud on my own server.
My main setup is this: I have a server that runs my models and workspaces. It might use Ollama, vLLM, LM Studio, LiteLLM, or another OpenAI compatible endpoint. I want to connect to that server from the OpenWork desktop app on another computer, start agent work, close my laptop, and come back later to see the same sessions, logs, diffs, approvals, and progress.
Problem / goal
This is not just remote model access. I want the actual agent work to keep running on the server.
For example, I connect OpenWork to my private worker server. The server stores the sessions, worktrees, logs, automations, MCP servers, and approvals. A local model does most of the work because it is fast and cheap. A stronger model can plan, review, or check the local model’s work. If I close my laptop, the server keeps going. When I reconnect, I can pick up where I left off.
A lot of people in local model communities talk about this. Local models are getting good, but they need a better harness around them. They need structure, permissions, review, and good context handling.
Primary user(s)
OpenCode primitive alignment
This seems like it could build on existing pieces: OpenCode sessions, permission requests, providers and models, MCP servers, skills, plugins, and remote workers.
The new part would be an OpenWork workflow layer for model roles, such as executor model, planner model, reviewer model, fallback model, small local model for normal work, and larger model for checking important work.
Some useful workflows would be: local model does the work and a stronger model reviews it, stronger model makes the plan and local model executes it, local model works until something fails and then escalates, or reviewer model checks diffs before the task is marked done. Human approval should still apply where configured.
Alignment with VISION/PRINCIPLES/PRODUCT
This feels aligned with OpenWork being local first, cloud ready, and able to connect to remote workers.
It would let people run their own private OpenWork setup without needing hosted cloud workers. The server owns the work, and the desktop app becomes the control panel.
Testability
Run an OpenWork worker on a server.
Connect from the desktop app.
Add a local model provider like Ollama or vLLM.
Start an agent session.
Close the desktop app.
Reopen it and reconnect.
Confirm the session, logs, diffs, approvals, and progress are still there.
Then test a workflow where a local model makes a change and a reviewer model checks it.
Ready to build it yourself?
Yes
Additional context
The main idea is that the server should own the work, and the desktop app should just be the control panel. That way I can use my own hardware, keep things private, and still have a good GUI for agent work.