Why it matters
AI agents in Whippy can work seamlessly across channels. A single agent can message, call, or do both, depending on where it is used. This flexibility lets you maintain one unified set of instructions instead of managing separate agents for each communication type.
Key Concepts
Agent: A configurable AI assistant that can handle both messages and calls using one shared instruction set.
Agent Instructions: A text-based configuration field where you define the agent’s purpose, behavior, tone, and any channel-specific expectations.
Call Settings: Optional call-related configuration that shapes how the agent behaves during phone interactions.
Agent Editor: The workspace where you configure instructions, settings, data parsing, and channels.
Knowledge Base: Optional information the agent can use for accurate responses across messaging and calls.
Step-by-Step: Create and Configure an AI Agent
Open the Sparkles icon from the left navigation.
Click Agents to open the agent list.
Click Create agent in the top-right.
Select Create from scratch or choose a Template.
Enter the agent name, status (Draft or Active), and description.
Click Save agent to open the Agent Editor.
Update Basic Info if needed.
Open Agent Instructions and define the agent’s purpose, tone, tasks, and communication style.
Add notes for how the agent should behave in messages.
Add notes for how the agent should behave on calls.
Emphasize any differences in phrasing or handling based on channel.
Review optional Call Instructions and adjust if the agent will be making or receiving calls.
Select the model, temperature, and tools as needed.
Click Save.
Configure Agent Settings: language mode, voice settings, speech-to-text preferences, responsiveness, interruptions, reminders, voicemail handling, and other call behaviors.
Save changes.
Configure Data Parsing and Analysis if you want structured information extracted from interactions.
Attach Knowledge Bases if needed, then save.
When ready, activate the agent and assign it to a channel such as a phone number or inbox.
Use Preview agent to test both chat and call behavior.
Tips and Best Practices
Keep instructions concise so the agent consistently follows them across modes.
Define messaging and calling behaviors separately inside the instruction text when needed.
Start with a lower temperature for more stable and predictable responses.
Test using chat first, then verify call behavior.
Use Draft mode while refining instructions, then switch to Active mode for production use.
Troubleshooting
Issue | Possible Cause | Fix |
Agent responses sound better in chat than in calls | Missing call-specific guidance in Instructions | Add explicit call behavior notes to the Instruction field |
Agent cannot take inbound calls | Channel not assigned | Activate the agent and assign it in Inbound Call Control |
Call audio or transcriptions are low quality | Missing boosted keywords or voice settings | Update vocabulary, voice settings, or boosted keywords |
Data extraction is incomplete | Parsing fields not defined clearly | Review and refine fields, labels, and examples in Data Parsing |