Set Up In-Conversation Tools
A voice agent that can only talk is half a teammate. Tools let it act while the call is still live — transfer to a human, navigate an IVR with keypad tones, book an appointment, hit an API for a lookup, or end the call cleanly. Build each tool once at the workspace level, then attach it to any agent that needs it.
Entry pointAutomation → AI Assistants → In-Conversation Tools
| Article summary | Details |
|---|---|
| Requires | Permission to manage tools and edit agents. |
| Where to go | Automation → AI Assistants → Agents → In-Conversation Tools (to create), or Build → Agent Details → Call Tools card (to attach). |
Tool Types
The Tool Type dropdown on the create form lists every type Brixi supports.
| Tool | Use it for |
|---|---|
| Call Transfer | Transfer the live call to a phone number or team. |
| Press Digit (IVR Navigation) | Send keypad tones (DTMF) so the agent can work through an automated phone menu — for example, "press 1 for sales" — when the number it dials is answered by an IVR. |
| End Call | End the phone call cleanly when the process is complete. |
| API Tool (HTTP) | Call an internal or external HTTP endpoint mid-call — lookup, status check, create or update a record. |
| Voicemail | Leave or handle a voicemail when the call isn't answered live. |
| Route To | Hand the conversation to another agent that is better suited to continue it (for example, route a buyer and a seller to different specialist agents). |
| Appointment Scheduler | Book an appointment or visit while speaking with the customer. |
| User Handoff | Hand the conversation to a human owner or team — used when AI should stop driving it. |
| End Conversation | Close the conversation path when the agent should stop engaging (without necessarily ending the phone line). |

Create an In-Conversation Tool
- Open Automation → AI Assistants → Agents and switch to the In-Conversation Tools tab.
- Select + Create Tool in the top-right. The create form opens under Back to Tools.
- Fill the Basics section:
- Tool Type — pick one of the tool types above. The Configuration section below changes to match the type.
- Name — the machine-readable name (snake_case is fine, for example
call_transfer). This is what the LLM sees in the prompt. - Display Name — the human-readable label shown in lists.
- Description — when the agent should use this tool, in plain English. The LLM reads this to decide whether to invoke the tool.
- Fill the Configuration section. The fields depend on the tool type — see the examples below for Call Transfer and API Tool (HTTP).
- Select Save at the bottom of the form. The tool appears under the In-Conversation Tools list.

Write the Description in agent-facing language. "When the customer asks to speak to a human, transfer the call" works better than "trigger transfer endpoint". The LLM uses this string to decide when to call the tool.
Attach a Tool to an Agent
A tool only runs when it's attached to an agent.
- Open the agent at Automation → AI Assistants → Agents → [agent name].
- On the Build → Agent Details view, find the Call Tools card on the right rail. Its subtitle reads "Actions available during calls."
- Select the + button on the card. A picker opens.
- Choose the tool you created. It moves into the card's attached list.
- Select Save in the agent header.

Update the prompt at the same time. The agent will only invoke a tool when the prompt explains the trigger condition. A tool attached without matching prompt rules tends to either never run or run too often.
Remove a Tool from an Agent
- Open the agent and expand the Call Tools card on Agent Details.
- Find the tool row and select the remove (×) control next to it.
- Select Save.
The tool stays in the workspace's In-Conversation Tools list — it's only detached from this agent.
Example: Call Transfer
A qualification agent should transfer to a sales owner when a customer is ready to buy or asks for a human.
Tool configuration
- Tool Type — Call Transfer.
- Name —
call_transfer. - Display Name — Call Transfer.
- Description — When the customer asks to speak to a human, asks for negotiation, or signals strong intent (e.g. "I want to book today"), transfer the call to the assigned sales owner.
- Configuration — the destination number or team and any fallback behavior if the transfer fails.
Prompt rule to add
If the customer says they want to speak to a person, asks for negotiation, asks for a custom payment, or says they want to book today, say "I'll connect you to a teammate now" and call the
call_transfertool.
Example: API Tool (HTTP)
Use an API tool when the agent needs to look something up mid-call — order status, the next available slot, eligibility.
Tool configuration
- Tool Type — API Tool (HTTP).
- Name —
check_order_status. - Display Name — Check Order Status.
- Description — When the customer asks about the status of their order, call this tool with the order ID they provide.
- Configuration — method (
GET/POST), URL, headers, body, and the Credential to use (see Credentials). The response is returned to the agent as an Action Result the prompt can reference.
Do not point an API tool at a production endpoint until you have tested it with safe data. A misconfigured tool that creates, updates, or deletes records will do so on every matching call.
Tool Design Rules
- One tool, one action. Don't fold transfer and end-call into the same tool.
- Name tools in business language. The LLM reads the name and description.
- Keep handoff and transfer destinations easy to audit — a real phone number beats a generic alias.
- Match prompt rules to the tool's description. The two have to agree.
- Test every tool from the agent's Test session before deploying.
Troubleshooting
| Problem | What to check |
|---|---|
| Tool doesn't appear in agent behavior | Confirm the tool is attached on Build → Agent Details → Call Tools and the prompt has a rule for it. |
| Transfer goes to the wrong number | Re-open the tool, check the destination in Configuration, and confirm the agent's phone number routing. |
| Appointment booking fails | Confirm required fields, calendar setup, and team availability. |
| API call fails | Check the credential selection, required parameters, and that the endpoint is reachable from Brixi. |
| Agent uses a tool too often | Tighten the prompt rule and the tool's Description. |
| Agent never invokes a tool that should run | Make the Description more specific and add a clearer trigger to the prompt. |
If the In-Conversation Tools tab or + Create Tool is missing, your role lacks tool management — ask your admin. Misfiring tools usually trace back to the prompt, not the platform: the agent only invokes a tool when the prompt spells out the trigger, so check the troubleshooting table before escalating.
Contact Brixi support if a transfer connects to the right destination in testing but drops or misroutes on live calls, or an API tool fails even though the same request succeeds outside Brixi. Include the tool name, the agent ID, and the time of a failed call so the attempt can be traced.