Repository navigation
Replies: 5 comments 1 reply
|
+1, I'd use this. One addition: it would be nice to choose how often titles refresh in settings, for example every X minutes, hourly, daily or weekly, along with the on/off switch. That way long threads keep a current, easy-to-understand title without me renaming them by hand. |
|
From a philosophical side of this: one chat - one goal. But I want to be honest, a chat's title gets degraded pretty fast, cause I cannot achieve initial goal in 50% of my chats yet. Maybe T3 needs something more than just a title? |
|
I’d like an opt-in setting to run T3 Code’s configured title generator after every completed user–agent exchange, once the agent’s final response has been saved. The goal is to keep the title relevant to the thread’s current topic and direction as the conversation evolves. The generation should include the latest user message and the completed agent response. The timing matters: asking the agent to call a regenerate-title tool cannot fully achieve this, because tool calls happen before the final response. T3 Code should trigger generation automatically after the turn finishes, independently of agent instructions or actions. Unlike the agent-driven approach described above, this would reuse the configured title generator on every completed exchange. Making it optional would let users choose whether the fresher titles are worth the additional generation calls. |
|
+1, specifically for a per-turn cadence rather than time-based. A setting like "regenerate the title every N completed turns" (N configurable, off by default) that runs the configured title generation model, not the thread's agent. That keeps it cheap (a small, low-effort title model costs almost nothing per call), works the same on every provider, and adds no timers or background polling. Manual renames should stay protected, as #10720 already does. |

Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Long-running threads can outgrow their initial title. Would opt-in automatic title updates be useful when the thread’s objective materially changes?
I’m testing this in a fork, built on the unmerged #10720. The link is to a source branch; there isn’t a packaged test release.
The agent working in the thread can update its generated title using the context it already has. No separate title-generation call is needed. Manually chosen titles stay protected, initial naming remains unchanged, and age, activity and cooldown requirements limit churn. Snoozed, archived and settled threads are excluded.
This came up in #7333, whose author primarily wants explicit renaming from a skill. Automatic updates are a separate, optional policy. An upstream contribution would aim to build on the rename/regenerate tooling in #2829.
The fork includes history and undo, which could remain a separate contribution. QA is ongoing and testers are welcome. Is this behavior something maintainers would want to explore?
Implemented with GPT-6 Astra in Codex. Reviewed with Claude Fable 5.1 and independently with GPT-6 Astra.
All reactions