robust-apricot•3mo ago
Can agents go backwards in flows
I have multiple agents in a network, each handling one task. The scenario is the user is at the warranty agent, but the conversation determined that they need to talk to the booking agent now. Can the user be directed back in the directional flow to then return to the booking agent? If so, how do we do this? Prompt or connections?
5 Replies
Yes – you can design your Voiceflow project to switch between agents based on the conversation’s current needs. While the flows are directional, you aren’t limited to moving only forward. Instead, you can use triggers (or connection steps) to "jump" from one agent to another.
For example, in your scenario:
• When the warranty agent detects that the conversation needs to switch to booking, you can have a trigger or action that directs the flow to the booking agent.
• You can also include a prompt that informs the user of the change before the trigger takes effect. This approach isn’t “going backwards” in the traditional sense; rather, it’s about re-routing the flow dynamically based on context and conditions. You can chain multiple agent steps together seamlessly by using logic and triggers, ensuring that the conversation path meets the user’s needs. For more details on chaining agents and using triggers, check out our documentation here: https://docs.voiceflow.com
• You can also include a prompt that informs the user of the change before the trigger takes effect. This approach isn’t “going backwards” in the traditional sense; rather, it’s about re-routing the flow dynamically based on context and conditions. You can chain multiple agent steps together seamlessly by using logic and triggers, ensuring that the conversation path meets the user’s needs. For more details on chaining agents and using triggers, check out our documentation here: https://docs.voiceflow.com
I'd probably just add a path in each agent, to go to a different agent. You'd add instructions like: "Trigger this path when the user needs to be directed to the Warrant team"
like @Brandersnatcher suggested you can add intent classification in each agent so the user can go to any of the other agents.
robust-apricotOP•3mo ago
Thanks for your help on this - by opening up the paths and creating logical intent prompts, the agents can navigate the agent network more effectively. This created a much more robust user experience!
I'd like to add to this and ask some advice. I have a 3 agent chain that does a survey, saving answers to variables at the end of each agent. The survey is too long for 1 agent cos they start to forget earlier answers if it's too long.
Agent 1 asks a few Qs and then routes to agent 2 for the rest, then agent 3. It took a while to get the prompt right in agents 2 and 3 and it seems a bit inconsistent. Also if the agent speaks another language, agent 1 won't move to agent 2 and I've no idea why. On the debug screen I can see agent 2 highlighted but it doesn't say anything.
When I manually ask it to continue, it does so but I shouldn't need to surely? Anyone have any specific things they found to work?