Tool Calls - telli API Documentation

Tool calls let your agent perform specific actions during a call — like ending the conversation, scheduling a callback, waiting for the customer, or calling an external API.

Quick Start

Reference a tool in your prompt

Use square brackets to reference tools: [toolName]

Define when to use it

In your system prompt, clearly explain when the agent should (and shouldn’t) use the tool

The agent handles the rest

During calls, the agent will trigger the tool when appropriate — the customer never hears the tool name

In the prompt, define when the agent should use a tool — not how the tool works technically. The technical configuration is handled separately in the agent settings.

How Tool Calls Work
Built-in Tools
Custom HTTP Tools

The agent decides — Based on your prompt instructions, the LLM determines a tool should be called
The tool executes — The tool runs in the background; the customer never sees or hears the tool name
The agent continues — The agent receives the result and continues the conversation naturally

These tools are available to every agent out of the box:

Tool	What It Does
`[endCall]`	Ends the call
`[callMeLater]`	Schedules a callback at a later time
`[waitForUserToReturn]`	Pauses and waits for the customer to come back
`[searchKnowledgeBase]`	Searches the agent’s knowledge base

Best Practices for Prompting Tool Usage
Example Usage Rules

Define When to UseFor each tool, clearly define in the system prompt:→ When the agent should use it→ When the agent should not use it→ What to say to the customer before/after using it (if anything)

Never Speak Tool Names Out LoudInstructions in square brackets are internal actions. The customer only hears the natural-language result.

Don’t say: “I’m now calling the searchKnowledgeBase tool”The customer shouldn’t hear technical tool names.

Do say: “Let me quickly look that up for you…”Keep it natural and conversational.

Place Instructions Close to UsageIf a tool is related to a specific section of your script (e.g., knowledge base search is related to troubleshooting), put the usage instructions in that section — not buried in a generic rules list far away.

Give Concrete ExamplesDon’t just say “use the knowledge base with Artikelnummer.” Show the agent exactly what a correct query looks like.

Don’t write: “Use Artikelnummer when in the installation flow”Too vague — the agent doesn’t know the exact format.

Do write: “When the customer is in the installation flow, always include the article number in your search query. Example: fulltext_query: 'Artikelnummer 000012345 Sky Stream Installation kein Bild'”Clear, specific, with a concrete example.

Define Negative CasesFor every “when to use,” also define “when NOT to use.” This prevents the agent from over-using or mis-using tools.

Tool	Action	Usage Rules	Example Situation
`[endCall]`	Agent ends the call	Only after all tasks are completed and the conversation is fully concluded.	The right person is not available
`[callMeLater]`	Agent schedules a callback	Always ask the customer first if they want a callback. Never call the tool before asking. Never use for expert callbacks. If the words “expert,” “colleague,” or “transfer” were used, this tool is forbidden.	The user is busy and wants to be called back later
`[waitForUserToReturn]`	Waits for the customer	Use when giving instructions and waiting for the customer to act, or when the customer asks you to wait.	Agent: “Plug in the cable.” User: “One moment, I’ll be right back.”
`[searchKnowledgeBase]`	Searches knowledge base	Use before giving technical instructions. Never use after already giving instructions.	Customer asks a product question not covered in the prompt

How to Prompt

​Quick Start

Quick Start