The chat loop

Every message you send that does not start with a slash is handled the same way underneath: it goes to tAI along with the context of your project, and the model works through it in a loop of its own tool calls until it has enough to give you a real answer.

One turn, many model responses

A single turn, everything that happens between you pressing enter and getting control of the prompt back, can involve many separate model responses chained together.

A safety cap on tool calls

There is a cap on how many tool calls a single turn can make, a safety limit rather than something you will normally notice, that exists so a model that gets stuck in an unproductive loop stops and reports back instead of running indefinitely.

Everything visible as it happens

Everything the loop does is visible as it happens, not just summarized afterward.

Managing long conversations

Long conversations accumulate context the same way any chat does, and at some point that context becomes large enough to slow things down or crowd out newer information.

What actually ends a turn

What ends a single turn is not a fixed number of tool calls, it is the model itself deciding it has done what your message asked for.

How tool results feed back into reasoning

Tool results feed directly back into the same reasoning that decided to call the tool in the first place, rather than being appended as an afterthought.

The same loop for remote-triggered turns

This same loop, unchanged, is what runs behind a remote-triggered turn sent from the tAI web app through a paired machine (see Pairing a machine).

Only one turn runs at a time

Only one turn ever runs at a time on a given machine, regardless of whether it was started locally or remotely.