Message queue in chat
You can send messages to a queue in the chat without waiting for the agent's previous response to fully complete.
How it works
While a message is being processed, a new message sent via the send button is added to the queue. The agent processes such messages sequentially, so none of them are lost.
Managing the queue
Send a message ahead of the queue
Use Alt+Enter when a new message should not wait for the requests already queued, but there is no need to stop the agent's current work. For example, you can immediately ask an independent question while the agent is doing a long task.
Shift+Enter or Ctrl+Enter add a new line to the message.
Open the queue
Click the button to the left of the send button to check the accumulated messages before a long agent run. In the queue window you can change the order, set the number of repeats, and remove outdated instructions.
Reorder messages
Drag a message to the position you want if the next steps depend on each other. For example, place the verification after the message with the code change, even if you wrote the verification first.
Repeat a message
Set the number of repeats if the model may need several consecutive attempts. For example, the message "Keep going until you are confident in the result" can be repeated several times for a weaker model. Choose the number of repeats with request cost in mind, and do not use repeats for actions that cannot safely be performed more than once.
Force-send a message
Use Force send when the current execution needs to be stopped: for example, the agent went down the wrong path, or a critical clarification came up. Hover over a message in the queue and click the force-send button. Explyt stops the current processing and sends the selected message.
Delete a message
Delete a message if it became outdated after the previous response or no longer matches the goal of the task. Hover over the message and click the delete button: it disappears from the queue and is not sent to the agent.
When it is useful
The queue is handy in a long iterative conversation, when you formulate the next steps before the agent's current response is finished.