Skip to main content

Prompts library

The Prompts library is your catalogue of reusable system-prompt templates. Write the prompt once, save it in the library with placeholder variables, and pull it onto any assistant — same shape, different values. It turns one well-tuned prompt into a fleet of assistants that share the same brand voice and the same constraints.

It's also where the shipped industry-specific starter prompts live (dental clinic, food ordering, survey) so you don't have to write from scratch.

What's in a prompt​

FieldWhat it is
NameInternal label — e.g. Sales SDR v3.
DescriptionFree text so teammates know what the template is for.
Template bodyThe system-prompt content, with {{placeholder}} variables for the parts that change per assistant.
VariablesLinked entries from your Variables library that fill in at runtime.

Template variables​

The grammar is small and substitution-only — no if, no loops. Five forms:

FormResolves to
{{contact.first_name}}Field on the matched contact (from caller phone or chat email).
{{<your_variable>}}Entry from your library (e.g. {{business_name}}, {{office_hours}}).
{{<provider>.<key>}}Value from a connected CRM — e.g. {{hubspot.lifecycle_stage}}.
{{<dynamic_key>}}Value passed in when you start the conversation via API.
{system.date_yyyy_mm_dd}System date/time — note single braces for system keys.

Other system keys: {system.date_time_utc}, {system.time_hh_mm}, {system.time_HH_mm}.

Unresolved placeholders are stripped from the rendered prompt — so missing values don't leak as literal {{...}} into the assistant's replies.

Three shipped starter templates​

Insighto ships with three system-owned prompt presets you can clone as a starting point.

Food ordering — phone​

For a restaurant or takeout-counter voice assistant. Asks for the order, confirms items and any allergens, and reads back the total before booking pickup time.

Dental clinic — phone​

For a dental front desk. Greets the caller, classifies new-patient vs. follow-up, collects insurance details, and books an appointment.

Samsung TV survey — phone​

A short post-purchase survey collecting four ratings and one open-ended comment. A useful pattern for any structured-questionnaire voice call.

You'll find these under Build → Library → Prompts with the system badge.

Four business use cases​

Agency — one template, twelve clients. A digital-marketing agency builds a Local Service SDR v2 template with {{business_name}}, {{services_offered}}, and {{primary_contact_phone}} as variables. For each of their twelve client assistants, they fill in the three variables. When they tune the master template, every client's prompt structure stays consistent.

Dental group — three locations. Bright Smile Dental clones the shipped Dental clinic template once, customises the body for their brand voice, and stamps three location-specific assistants from it (each with its own {{office_address}}, {{office_hours}}, and Calendar Pool).

E-commerce — seasonal swaps. A meal-kit company keeps one canonical support prompt with a {{current_promo}} variable. Every Monday morning they update the variable value with that week's promo code; every assistant linked to the prompt template picks up the change.

B2B SaaS — quick A/B tests. An analytics startup keeps Sales SDR v3 (canonical) and Sales SDR v4 (testing) as siblings in the library. They point a staging assistant at v4, compare results in Conversations for a week, and rename if the test wins.

Creating a prompt template​

  1. Open Build → Library → Prompts → New prompt.
  2. Set a name and description.
  3. Write the template body. Use {{variable_name}} for the parts that should change per assistant.
  4. Under Variables, link any entries from your Variables library that this template uses.
  5. Save.

The template is now available to copy onto any assistant in the same workspace.

Applying a prompt to an assistant​

Open the assistant, click System prompt → Use template, pick the template, and Insighto renders the body with the assistant's linked variable values into the assistant's system prompt field.

Two ways to use it:

  • Copy and detach — paste the rendered template into the assistant's system prompt directly. Future edits to the library template don't flow through automatically.
  • Keep aligned by convention — point multiple assistants at the same set of variables and re-apply the template when you tune it.

There is no automatic "linked forever" relationship today — re-apply when you update the template.

Example template body​

You are {{assistant_role}} for {{business_name}}.

Your job is to greet the caller, identify why they are calling,
and either book an appointment or transfer to a human.

Hours: {{office_hours}}
Address: {{office_address}}
Today is {system.date_yyyy_mm_dd}.

Tone: warm, concise, never robotic.

When the caller wants to book, collect their name, phone, and
preferred slot, then call book_appointment.

If the caller asks something outside scope, say "Let me get
someone to help with that" and call transfer_to_human.

{{contact.first_name}} — say their name once if available.

{{assistant_role}}, {{business_name}}, {{office_hours}}, {{office_address}} come from variables linked to the template. {{contact.first_name}} is filled in per call from the matched contact. {system.date_yyyy_mm_dd} is always present.

Versioning convention​

There's no built-in revision history. A light convention works:

  1. Keep one canonical template per role (e.g. Sales SDR v3).
  2. Before a substantive change, clone to Sales SDR v4 (testing).
  3. Point a staging assistant at v4.
  4. Compare in Conversations for a week.
  5. Rename v4 to canonical and retire v3 when the test wins.

Where to next​