Bug: SSE /event stream closes immediately after server.connected event

Description

Description:

The opencode server's event stream (GET /event) terminates the connection immediately after sending the initial server.connected event. No subsequent bus events (e.g., message.part.delta, message.part.updated) are received by the client, despite the server logs showing that these events are being published internally.

Environment:

OS: Linux Ubuntu 24.04
OpenCode version: 1.14.46 (Broken) / 1.14.41 (Working)
Install method: curl -fsSL https://opencode.ai/install | bash

Steps to reproduce:

Run opencode serve.
Connect to the event stream using curl -sN http://localhost:4096/event.
In another terminal, send a prompt to a session via the HTTP API (e.g., using POST /session/:id/message). (or simply opencode attach http://localhost:4096 and send something)
Observe that the curl connection closes immediately after the server.connected event, and no message events are streamed.

Verification:

Server Logs: opencode.log confirms that events are being published to the internal bus (e.g., service=bus type=message.part.delta publishing).
Direct API Test: curl -vN http://localhost:4096/event returns HTTP 200 OK and the first event, but then the connection is closed by the server.
Version Regression: The issue is present in v1.14.46 but is completely resolved when rolling back to v1.14.41.

Likely cause:

The regression appears to have been introduced between v1.14.41 and v1.14.46. A strong suspect is the introduction of HTTP API response compression in v1.14.42, which may be incorrectly buffering or terminating the long-lived SSE stream.

OpenCode version

1.14.46

Operating System

Ubuntu 24.04

Terminal

WezTerm

This issue might be a duplicate of existing issues. Please check:

prompt_async silently discards requests in v1.14.44 (SSE path broken) #26635: Same symptom — SSE /event stream only receives server.connected and then closes with no subsequent events (message.part.delta, etc.) after using prompt_async. That issue also identifies this as a regression starting in v1.14.44+, which overlaps with your reported broken version (v1.14.46).

I found that v1.14.41 works, but v1.14.42 is already broken. This narrows the regression window down to the changes introduced in v1.14.42.

fix(server): recover OpenCode turns when 1.14.42+ SSE drops early (#902)

This affect my opencode plugin, It does not work anymore when streaming new messages.

sudo-tee/opencode.nvim#374

I’m experiencing the same issue. v1.14.41 works, but newer versions up to 1.14.48 are broken.

I am experiencing this issue too while testing Kimaki and making my own lightweight Discord bridge. It seems the /event endpoint is no longer used by first-party user interface.

Instead, both the TUI and the Web app now uses the /global/event endpoint. Migrating the app to the new SDK method sdk.global.event worked around this issue.

I am experiencing this issue too while testing Kimaki and making my own lightweight Discord bridge. It seems the /event endpoint is no longer used by first-party user interface.

Instead, both the TUI and the Web app now uses the /global/event endpoint. Migrating the app to the new SDK method sdk.global.event worked around this issue.

Thanks for the info. Unfortunately the /global/event does not fix our issue, as the messages do not contain the text property, so we still cannot stream the events to the buffer with it.

@lokikl

Adding another data point: this regression also affects in-process plugin event delivery (via bus.subscribeAll()), not just the external SSE /event stream.

I have a server-side plugin that subscribes to bus events to drive a status reporter. Bisecting between v1.14.41 (works) and v1.14.42 (broken) — identical to @lokikl's bisect — I captured exactly which events the plugin's event handler receives:

v1.14.41 — outer event types seen by bus.subscribeAll():

message.part.delta
message.part.updated
message.updated
session.created
session.diff, session.idle, session.status, session.updated, tui.toast.show

v1.14.42 — outer event types seen by bus.subscribeAll():

message.part.delta
session.diff, session.idle, session.status, session.updated, tui.toast.show

Missing in v1.14.42: message.part.updated, message.updated, session.created.

The session data is still correctly persisted — GET /session/{id}/message returns the full assistant message with step-start, text, step-finish parts — but the bus never announces those parts to plugin subscribers. Net effect: any plugin that drives state machines off message.part.updated (step-start / text / reasoning / tool / step-finish) stops working silently.

#26936 ("fix(tui): scope events by project", merged into dev today) appears to address the TUI side via a new event-subscription API with metadata, but it doesn't change bus.subscribeAll(), which is what plugins (and any non-TUI subscriber) use. Just flagging so the in-process plugin case isn't missed when the fix lands.

@angelolloqui

I ran into this issue with my emacs opencode integration: https://codeberg.org/sczi/opencode.el
I switched to subscribing to the /global/event endpoint and everything is working again. @angelolloqui in 1.14.48 with the /global/event endpoint I do receive session.created, message.updated, etc events.

Confirming this matches our openchamber/openchamber#1258 and #27319.

On Arch Linux with bun 1.3.13, the 1.14.41 binary unfortunately crashes on startup (segfault in bun executable — different issue from the SSE bug). This makes the clean rollback inaccessible for us.

The HTTP API compression theory (introduced v1.14.42) is compelling — would be great if that could be verified/toggled independently of a full version change.

I also switched to the global/event endpoint and normalized the events so they match and it resolved the issue in the meantime

Thanks for the suggestion! I will try that out. Is this change documented somewhere? Such a breaking change should probably be well explained if it was intended

this wasn't intended we're looking into it now

fixed: b928a1f

fix(server): recover OpenCode turns when 1.14.42+ SSE drops early (#902)

fix(server): recover OpenCode turns when 1.14.42+ SSE drops early (ge…

Navigation Menu

Repository navigation

Bug: SSE /event stream closes immediately after server.connected event #26697

Description

Description:

Environment:

Steps to reproduce:

Verification:

Likely cause:

OpenCode version

Operating System

Terminal

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Bug: SSE /event stream closes immediately after server.connected event #26697

Description

Description

Description:

Environment:

Steps to reproduce:

Verification:

Likely cause:

OpenCode version

Operating System

Terminal

Activity

github-actions commented on May 10, 2026

lokikl commented on May 10, 2026

sudo-tee commented on May 11, 2026

KryptXBSA commented on May 11, 2026

dtinth commented on May 12, 2026

sudo-tee commented on May 12, 2026

angelolloqui commented on May 12, 2026

sczi commented on May 13, 2026

Canahmetozguven commented on May 13, 2026

sudo-tee commented on May 13, 2026

angelolloqui commented on May 13, 2026

thdxr commented on May 14, 2026

rekram1-node commented on May 14, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions