Why it matters
AI agents are the core of automation in Whippy. A single agent can handle conversations across messaging and calling, as long as it is configured correctly.
The most important part of any agent is its instructions. Agent Instructions control how the agent behaves, when it references knowledge bases, and when it uses action tools. Knowledge bases and tools expand capability, but instructions determine how everything operates together.
Without clear instructions, an agent may respond inconsistently, ignore tools, or misuse knowledge.
Key Concepts
Agent: A configurable AI assistant that can operate across chat, voice, or both using one unified setup.
Agent Instructions: The core logic layer of the agent. Instructions define tone, behavior, priorities, decision-making, when to reference knowledge bases, and when to use tools. They determine how the agent operates in all scenarios.
Channels: Where an agent operates, such as messaging inboxes or phone numbers. Capabilities depend on channel assignment.
Knowledge Base: Optional reference material that provides factual information the agent is allowed to use.
Information Source Types: The four ways content can be added to a knowledge base: Text content, Document upload, Single webpage, and Website.
Multiple Knowledge Bases: An agent can reference more than one knowledge base at the same time.
No-Knowledge Agents: An agent can exist without any knowledge base attached and rely only on instructions and tools.
Agent Tools: Optional capabilities that allow the agent to take structured actions or retrieve information dynamically.
Action Tools: Tools created using the Execute Actions tool type. These allow agents to perform system actions such as sending messages, tagging contacts, assigning users, updating conversation status, managing sequences, sending emails, starting chat agents, or triggering HTTP requests.
Agent Settings: Controls that affect voice delivery, transcription, language handling, responsiveness, and call behavior.
Data Parsing: Optional configuration for extracting structured data from conversations.
Versions: A history of saved changes to an agent, allowing review and rollback.
The Role of Agent Instructions
Agent Instructions are the foundation of how an agent operates.
They define:
How the agent speaks and behaves.
What its purpose is.
What it should prioritize.
When it should reference knowledge bases.
When it should trigger action tools.
What conditions must be met before taking an action.
Knowledge bases do not automatically control responses. Tools do not automatically execute. Instructions decide if and when either is used.
If instructions do not explicitly or implicitly guide the agent to use knowledge or tools, the agent may ignore them.
Instructions are not optional configuration. They are the control system for the entire agent.
Step-by-Step: How AI Agents Work in Whippy
Create an agent with a name and description.
Define Agent Instructions. This step determines how the agent behaves and operates.
Optionally attach one or more Knowledge Bases to provide factual information.
Optionally configure Tools, including Action Tools, that the agent can use.
Assign the agent to one or more Channels, chat, phone, or both.
Publish or activate the agent.
During conversations, the agent:
Uses its instructions to interpret user intent.
Decides whether to answer directly or reference a knowledge base.
Determines whether a tool should be triggered.
Executes actions only when conditions defined in instructions are met.
Instructions are evaluated first. Knowledge and tools are secondary layers that depend on those instructions.
Step-by-Step: Explore Agent Configuration
Click the Sparkles icon from the left navigation.
Select Agents to view your agent list.
Open an agent to access the Agent Editor.
Review Basic Info for name, description, and status.
Open Agent Instructions. Confirm behavior rules, tool usage logic, and knowledge guidance are clearly defined.
Check for channel-specific guidance for chat and calls.
Review Agent Settings for voice and transcription behavior.
Open Data Parsing and Analysis to see structured data extraction.
Review Knowledge Bases attached to the agent, if any.
Review Tools, including Action Tools, to understand what actions the agent can perform.
Check Versions to review past changes.
Knowledge Bases and Information Source Types
Knowledge bases act as the agent’s factual reference layer. They do not control behavior.
Each knowledge base is made up of pages, and each page uses one information source type.
Text Content
Manually written or pasted information.
Used for FAQs, policies, scripts, and structured explanations.
Document Upload
Files such as PDFs, Word documents, spreadsheets, or CSVs.
Used for handbooks, SOPs, training materials, or compliance documentation.
Single Webpage
A single public webpage URL.
Content is scraped at publish time.
Used for specific pages such as location details or pricing pages.
Website
A website URL with include and exclude path rules.
Controls which sections of a site are ingested.
Used for help centers, FAQs, or multi-location information such as opening hours.
Only published knowledge base content is available to agents.
Instructions must guide the agent on when to reference knowledge.
Agent Instructions vs Knowledge Bases vs Action Tools
Each component has a separate role.
Agent Instructions
Control logic, behavior, and decision-making.
Define when knowledge should be referenced.
Define when tools should be used.
Act as the operational core of the agent.
Knowledge Bases
Provide factual information.
Do not define behavior.
Are used only if instructions direct the agent to rely on them.
Action Tools
Allow structured system actions.
Do not run automatically.
Require explicit or clearly defined trigger logic in instructions.
Instructions determine:
Whether knowledge is used.
Whether tools are triggered.
What conditions must be met before acting.
Without strong instructions, adding knowledge bases or tools will not produce reliable automation.
Example: Receptionist Agent Using Knowledge and Action Tools
A receptionist agent handles calls and messages.
Agent instructions define:
Greeting structure.
Escalation rules.
When to reference opening hours.
When to assign a user or notify staff.
A Website knowledge base provides location-specific hours.
An Assign User or Notify Users action tool alerts staff when a follow-up is required.
The instructions control when knowledge is consulted and when actions are triggered. The knowledge base provides data. The tools execute tasks.
Tips and Best Practices
Write clear, structured Agent Instructions before attaching knowledge or tools.
Define exact conditions for tool usage.
Specify when knowledge must be referenced.
Avoid relying on implied logic. Be explicit.
Test behavior after updating instructions.
Review transcripts to determine whether issues stem from instructions, knowledge, or tool logic.
Troubleshooting
Issue | Possible Cause | Fix |
Agent ignores tools | No tool usage logic in instructions | Add explicit trigger conditions in Agent Instructions |
Agent does not reference knowledge | Instructions do not direct knowledge usage | Add guidance to reference the knowledge base |
Agent performs actions unexpectedly | Trigger logic too broad | Refine and tighten instruction conditions |
Agent gives incorrect information | Knowledge base outdated | Update and publish the knowledge base |
Agent gives inconsistent responses | Instructions unclear or conflicting | Simplify and clarify instruction hierarchy |
Agent does not perform an action | Tool not assigned or no instruction logic | Enable the tool and define when to use it |
Agent behaves differently across channels | Missing channel-specific rules | Add explicit chat versus call guidance |