iostack.ai Portal

Transitions allow the author to specify possible paths between stages.

They are the mechanism by which the agent can move from one task to another, ie from one prompting context to another.

A transition provides a clear enough criterion to allow the agent to decide when to move from the current stage to another stage.

What’s in a transition?

A Transition contains:

A condition or reason to take the transition. The condition is defined by you, the author, in plain language.
A transition also contains an optional list of external integrations to trigger when executing the transition between stages.

The plain language specification of the condition for a transition to be taken is included as part of the prompting context presented to the LLM at each turn occurring in that stage, for all of its transitions.
This means that a transition condition could be as simple as ‘the user answered 'yes'` or ‘you have collected and validated x piece of data’.
It also means that it can also be extremely nuanced eg. ‘Using your experience and the interaction with the user, you are satisfied that they have understood the consequences of x in enough detail that they are ready to proceed’.

There can be multiple transitions for each stage, each specifying a path ‘out’ of the stage, allowing you to create branches in your conversational structure.
Similarly, there can be multiple transitions into a stage, representing different paths converging at the same stage or point in the conversation.
At each turn occurring in the stage, the LLM assesses each transition’s condition, choosing the most appropriate transition to take (if any) and thereby allowing it to navigate its own way from one stage to the next according to the content of the conversation with the user.

Just as for integrations in stages, integrations defined in transitions enable retrieving or sending data.
Integrations configured for a specific transition are triggered when the conversation flow takes that transition.