Pause AI, not Stop, not Steer, not Interject, just pause so we can keep up with "the train of thinking tokens", for those who care.

[dan-oak]

What are you proposing?

Pause button along with the already existing Stop button in AI chat.

i absolutely love Steer function that was recently added in #59310 by @danilo-leal !
it was exactly what i missed from the other harnesses that i used in the past (not sure if those already caught up with this feature or not).

Why does this matter?

the issue still that i want to raise: Pause is very important too, not only Steer. for us, who care about following the agent's investigation/research/reasoning flows, i often just want to pause it at whatever logical point is is about to hit, just to read along to the point where it is now. LLMs generate tokens fast, and mine current AI of choice (DeepSeek pro versions) dump a lot of thinking tokens. often, depends on a project, it goes into depths i dont need (it's way too eager too do things right and investigate to the root of the problem, if you give it all the resources). so i want to have an option to Pause.

currently i can just Stop. but then to continue i need to awkwardly type a dummy message like "please continue (i force-stopped your execution)" or, if i want to redirect the work, then something like "stop, let's look into this or that instead".

Possible approach

i do not have ideas on how to implement this, and i might suspect that there can be difficulties in integrating such feature with some AI providers, but maybe it is feasible.

[lafkpages]

+1.

I think this should be easy to implement if we allow for the current LLM request to finish streaming, but then pause before any more actions such as tool calls or file edits. This would make the feature work with all existing LLM providers while letting you read the LLM's thought process before it goes on and does anything else.