Feature hasn't been suggested before.
Describe the enhancement you want to request
OpenCode can keep a long-lived server process with cached instances for different projects/worktrees. When working on project-scoped configuration or extension files, such as agents, skills, plugins, or MCP-related setup, the current instance can keep using the old bootstrapped state.
The current workaround is usually to restart the server/app, or to dispose the instance and let a later request recreate it. Restarting is too broad because it affects unrelated projects/worktrees. Disposing is also not a great user-facing reload action because it only tears down the old instance; it does not tell the caller whether a replacement instance can boot successfully.
I would like a small manual reload action for the currently routed project/worktree instance.
A first version could add:
POST /instance/reload, scoped by the existing workspace/directory routing.- A typed API error when the replacement instance fails to boot.
- A WebUI
Reload instance action that shows success/failure feedback.
This is intentionally smaller than full hot reload. It does not need to watch files automatically, reload every workspace, reconcile MCP clients incrementally, or guarantee that active runs inside the same instance continue uninterrupted.
Feature hasn't been suggested before.
Describe the enhancement you want to request
OpenCode can keep a long-lived server process with cached instances for different projects/worktrees. When working on project-scoped configuration or extension files, such as agents, skills, plugins, or MCP-related setup, the current instance can keep using the old bootstrapped state.
The current workaround is usually to restart the server/app, or to dispose the instance and let a later request recreate it. Restarting is too broad because it affects unrelated projects/worktrees. Disposing is also not a great user-facing reload action because it only tears down the old instance; it does not tell the caller whether a replacement instance can boot successfully.
I would like a small manual reload action for the currently routed project/worktree instance.
A first version could add:
POST /instance/reload, scoped by the existing workspace/directory routing.Reload instanceaction that shows success/failure feedback.This is intentionally smaller than full hot reload. It does not need to watch files automatically, reload every workspace, reconcile MCP clients incrementally, or guarantee that active runs inside the same instance continue uninterrupted.