AI agent flows and tools

Most Drop Cowboy® AI agents work from their prompt alone. Some need more control: a fixed call script, a way to pass callers on, or a custom tool that reaches your own systems.

Build a conversation flow

Do this when every call must follow the same steps in order, such as an intake that collects set details before anything else.

1. Open AI Agents in the left menu and click the agent. If you don't see AI Agents, ask your account owner for access.
2. Open the Flow tab. Under Call handling, turn on Guide the call with a fixed script.
3. Click Create a script. To reuse one you already built, pick it in Use an existing script instead.
4. The flow builder opens. Click a node to edit it in Node properties:
   • Display Name: what you call the step.
   • System Prompt / Task: what the agent does or says at this step.
   • Required variables under Required before leaving this step: details the caller must give first, such as a phone number or email address.
   • Tool to run and What to send the tool: for an Action node.
   • Routes: a Match (intent / keyword) and the Go to node it leads to. Set a Default next node for everything else.
5. To add a step, pick a type (Entry, Conversation, Action, Decision or End) and click Add node.
6. Click Validate and fix anything it lists. Click Save.
7. Click Publish. The status changes from Unpublished changes to Live. Click Back to agent.

To change it later, open the Flow tab and click Edit script. To stop using it, click Unlink.

If it doesn't work:

• The flow says "Generated from this agent's persona". It was built from the agent's prompt. Once you edit and save it, it stops following later prompt changes. You see "These edits will disconnect this flow from the persona" before you save.
• "Possible automation loop detected". A tool or step may start an automation that starts this call flow again. Read the message under the warning and change the step it names.
• Calls still follow the old steps. Check the status reads Live. If it reads Not published or Unpublished changes, click Publish.

Transfer a call to a person

Do this when the agent should pass a caller to you, your team, or another agent, for example when a caller asks for a person.

1. Open AI Agents in the left menu and click the agent. In Manage, click Re-run setup wizard, then click Transfers in the step bar. A new agent reaches this step during setup.
2. Under Set up transfers, click a card in Add a transfer:

1. Fill in Description for agent. This tells the agent when to use the transfer, for example "Caller asks for billing or a human".
2. Click Add transfer, or Add handoff for an agent. Add more transfers the same way.
3. Click Continue to Review & finish, then Finish Setup.

You can also add transfers on the agent's Capabilities tab, under the Routing pill. Use Hand off to another AI agent with Add handoff and When should it hand off?, or Transfer to a phone line with Phone Lines. Click Save changes.

To have the agent introduce the caller before connecting, open Advanced and set Transfer mode to Warm transfer (announce first). Cold transfer (blind) connects the caller with no introduction.

If Record calls transferred here is on, tell callers they may be recorded where the law requires it. Ask your legal counsel what to say.

If it doesn't work:

• "Transfers to phone lines require a verified account." Click Verify your account and finish verification, then add the transfer again.
• Another Phone Line has no lines to pick. Click Open Phone Hub and set one up. See Set up a phone line.
• The agent never transfers. Make Description for agent name the exact situations, then test it. See Test your AI agent.

Add a custom tool

Do this when the agent needs to reach your own systems mid-call, for example to look up an order or create a ticket.

1. Open AI Agents in the left menu and click the agent.
2. Open the Capabilities tab and the Custom Tools pill. Click Add tool.
3. Fill in the form:
   • Slug: a short name with letters, numbers and underscores, such as lookup_order.
   • Description: when the agent should use this tool. The agent decides from this text.
   • Parameters: a JSON Schema for what the tool needs. Leave it blank if it needs nothing.
   • Latency: Fast, or Long (plays a wait message) with a Wait message for the caller.
4. Pick a Binding type, which is what the tool runs:
   • Automation flow: pick the Flow and a Mode, Sync (wait for result) or Async (fire and continue). See Build an automation.
   • Piece action: pick a Piece, Action and Connection.
   • HTTP webhook: enter the URL, Method and any Authorization header.
   • MCP tool: enter the MCP server name and MCP tool name.
5. Check Can be retried automatically without side effects only if running it twice is harmless. Set Max invocations per call to cap how often it runs. Leave Enabled on.
6. Click Add tool. Custom tools save right away.
7. Click the test icon on the tool's row, fill in Sample arguments, and click Run test. A tool bound to an automation flow runs the real flow, so you first confirm with I understand, run it anyway.

If it doesn't work:

• Run test shows Failed. Read the error, fix the URL, connection or flow, and run it again.
• The agent never uses the tool. Rewrite Description to say exactly when to use it, then test the agent.
• Callers wait with nothing said. Set Latency to Long (plays a wait message) and add a Wait message.

This information is for educational purposes only and does not constitute legal advice. Regulations vary by jurisdiction and use case. Always consult with qualified legal counsel to ensure your specific practices comply with applicable federal and state laws.