Capture caller details with Extract Variables | FieldCamp

The Extract Variables step reads the conversation so far and saves the details you ask for, such as the service address or whether the caller says it's urgent. It doesn't speak to the caller. Once it has run, the values are available to every later step as variables.

What the step does

In Add a step, the step is described as "Silently extract typed values from the conversation before continuing." When the call reaches it:

1. It reads the conversation and fills in each variable you declared.
2. If the caller didn't give a value, or the value doesn't fit the type, that variable is set to null.
3. The call moves on to the next step without pausing.

An Extract Variables step has exactly one connection going out. When you connect it, the builder creates that connection for you, labelled "Continue", and it's silent. It also needs at least one connection coming in.

Add the step

Add a variable

Each variable has three fields:

• Variable Name: a snake_case name, such as service_address or is_emergency. This is the name you use in later steps. The names nodes_visited and logic_split_trace are reserved.
• Type: the kind of value to capture.
• Extraction Hint: what to look for in the conversation for this one variable.

Variable types

For an Enum, list the allowed choices. They're matched exactly, including capital letters, so keep them simple and consistent.

Capture phone numbers and ZIP codes as String, not Number, so leading zeros and formatting survive.

Write good hints

• Say exactly what counts. "Street address where the work is needed, including unit number" works better than "address".
• Say what to do when it's missing. An Extraction Prompt such as "Leave a value empty if the caller did not say it" stops the step from guessing.
• Use your own rules for judgement calls. For is_emergency, write down what your company treats as an emergency, for example "true only if the caller reports no heat with someone vulnerable in the home, or a gas smell". These are examples; replace them with your rules.
• Keep one fact per variable. Split "address and access notes" into service_address and access_notes.

Use extracted values

Captured values are stored under gathered_context. Use them anywhere variables work:

• In prompts of later steps: {{gathered_context.service_address}}, or the short form {{service_address}}. A captured value overrides a starting value with the same name.
• In a Logic Split rule: gathered_context.is_emergency == true.
• In a Webhook payload: "address": "{{gathered_context.service_address}}".
• In tool settings that accept variables.

If a value is null or was never captured, the variable becomes empty text. Add a fallback so the prompt still reads well, for example {{gathered_context.service_address | the address you gave}}. See Variables for the full syntax.

Example: new service request

A plumbing company wants to capture the job, then route emergencies differently.

1. An Agent step named "Ask about the job" asks what's wrong, where the property is, and the best callback number.
2. An Extract Variables step named "Capture job details" has this Extraction Prompt: "Pull the service address, the problem, and whether the caller says water is leaking right now. Leave a value empty if the caller did not say it." Its variables:

3. A Logic Split sends is_emergency == true to your emergency process and everything else to an End Call step that reads back {{gathered_context.service_address}} and {{gathered_context.callback_number | the number you called from}}.

If you've connected FieldCamp, you can save these details there during the call: give a later Agent step the FieldCamp App Create request tool; see Look up clients and jobs during a call. Your team then works the request as described in Requests & Pipeline Overview.

Frequently asked questions

What caller details should I capture?

Capture what your team needs to act on the call, usually the caller's name, service address, callback number, a short description of the problem, and anything you route on, such as urgency or service type. Keep each detail in its own variable.

Which variable types are supported?

Extract Variables supports String, Number, Boolean and Enum. Enum values must match one of the choices you list exactly, including capital letters.

What if the caller doesn't give a value?

The variable is set to null and the call carries on. In prompts it becomes empty text, so add a fallback such as {{service_address | the address you gave}}, and in a Logic Split the Else branch catches it.

How do I use an extracted value later in the call?

Write it as {{gathered_context.variable_name}} in a later prompt, Webhook payload or tool setting, or use the path gathered_context.variable_name in a Logic Split rule.

Is Extract Variables the same as post-call extraction?

No. Extract Variables runs during the call, so later steps can use the values. Post-call extraction, in the agent's Settings, runs after the call to produce a summary, outcome and other fields.

Related articles

• Branch a call with the Logic Split step
• Use variables in your voice agent
• Configure an Agent step
• Use API trigger, webhook, and QA steps