You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
[FEATURE]: Configuration option to disable message summary generation #6228
I have verified this feature I'm about to request hasn't been suggested before.
Describe the enhancement you want to request
Is it possible to opt-out of the additional summarization calls that occur per message? I was hoping to see a configuration option skip these requests.
In this example, I send 2 requests to devstral, and opencode inserts an additional 3 requests for summary generation to haiku. (this is a screenshot of Helicone, which I'm using to monitor requests from opencode)
A summary is requested for every user message. And an additional summary request generates the session title, which is essentially a duplicate of the first message summary request.
I see the session summary in the TUI, but don't see any message summaries. So the message summaries don't seem to be adding any value for TUI users.
The message summaries are used in the web/desktop app, but I'm not using this app, and even if I did, I would prefer to downgrade the message summaries to just the first few words of each message in order to avoid the additional summarization requests.
Could this be fixed? The summaries are a worthwhile feature, I think (in particular, the diffs on the web interface), however, I'm not sure the LLM generation is worth the extra time delay when running local models. These should be separately configurable: the computed diff, the per-message summary, and the per-session title.
Activity
milesfrain commentedon Dec 27, 2025
A workaround to prevent all summaries is to set
small_modelto a non-existent model, like this:https://opencode.ai/docs/config/#models
Although it would be nice to keep some summaries, like the session summary.
Nindaleth commentedon Dec 27, 2025
Another, hopefully more targeted, possible workaround is to disable the title agent:
However that, too, will result in a generic session name not reflecting the initial user prompt.
thdxr commentedon Mar 25, 2026
To stay organized issues are automatically closed after 90 days of no activity. If the issue is still relevant please open a new one.
SirNate0 commentedon Apr 16, 2026
Could this be fixed? The summaries are a worthwhile feature, I think (in particular, the diffs on the web interface), however, I'm not sure the LLM generation is worth the extra time delay when running local models. These should be separately configurable: the computed diff, the per-message summary, and the per-session title.
erikvanzijst commentedon Oct 4, 2026
Agreed, this is not a great fit for use with local models that have limited context parallelism.