Docs
Open dashboard
Get started

Introduction to Inspra

Inspra is a platform for building, testing, deploying and monitoring AI voice agents. By the end of the quickstart you will have an agent that answers or places calls, has been tested in the browser, and has its first calls recorded for review.

Choose your path

Most readers work in the dashboard. Developers who connect other systems also need the credentials and events that the dashboard manages. The two paths are kept separate so that each reader only follows the steps that apply.

How to use this guide

Follow the navigation path stated beneath each workflow’s introduction. Screenshots show the relevant controls; your own names, counts, connected providers and enabled options may differ from the illustrations, since they were captured in one example workspace. On this page, select a screenshot or its View full size link to inspect the controls in a new tab.

Choose the correct workspace before making changes. Save configuration changes before testing. Creating a campaign, starting a test, sending a message or activating an automation is an operational action and may use connected services or credits.

The first agent sequence

  • Define the conversation goal and boundaries.
  • Create the agent and choose its voice.
  • Write the prompt and attach approved knowledge.
  • Configure any required tools and analysis.
  • Test the agent and review the conversation.
  • Connect a number or create a reviewed campaign.
  • Monitor calls and improve the configuration.
Example. A first agent with one job: answer common product questions from an approved knowledge document, and transfer to a person for anything else. Test it with Web Call, then assign an inbound number.

Agent providers

Every agent is created on one provider, and the provider determines which voice, model and tool options the form shows. This documentation always presents the providers in the order below and walks through Native Agent first. The provider cannot be changed after the agent is created.

  1. 1Native Agent
  2. 2Vapi
  3. 3Retell

See Create an agent and choose a provider for the comparison and a separate setup path for each provider.

Build

An agent combines a provider, a voice, a first message, a prompt, approved knowledge and the tools it needs. Start from the form for direct control, or describe the job in plain language to Inspra Copilot, an AI assistant that prepares a configuration for your review.

Create an agent with the form

In Agents, select Create Agent. Name your agent and choose Native Agent, Vapi or Retell, then configure its direction, model and voice.

Read the guide : Create an agent with the form
Create an agent with the form in the dashboard
View full size: Create an agent with the form (opens in a new tab)

Inspra Copilot

Select Build with Composer in Agents, then describe the job. Review the proposed configuration before creating the agent.

Read the guide : Inspra Copilot
Inspra Copilot in the dashboard
View full size: Inspra Copilot (opens in a new tab)

Write the opening message and prompt

Open the agent's Prompt tab. Set who speaks first, write the greeting, and use the system prompt to define its scope and how it handles missing information.

Read the guide : Write the opening message and prompt
Write the opening message and prompt in the dashboard
View full size: Write the opening message and prompt (opens in a new tab)

Create a knowledge document

In Knowledge Base, select Add Document and choose Write Content. Add a title and approved information, then attach the document to your agent.

Read the guide : Create a knowledge document
Create a knowledge document in the dashboard
View full size: Create a knowledge document (opens in a new tab)

Test

Hear the agent before it reaches a customer. Browser calls check the greeting and behaviour; simulated scenarios and scorecards repeat the same checks after each change.

Test with a browser call

Speak to the agent from your browser, try an edge case and review the call record.

Read the guide : Test with a browser call
Test with a browser call in the dashboard
View full size: Test with a browser call (opens in a new tab)

Create a simulated test scenario

Describe a caller and the expected behaviour, then run it against the agent.

Read the guide : Create a simulated test scenario
Create a simulated test scenario in the dashboard
View full size: Create a simulated test scenario (opens in a new tab)

Configure call analysis and scorecards

Decide what each call is evaluated on before you rely on the results.

Read the guide : Configure call analysis and scorecards
Configure call analysis and scorecards in the dashboard
View full size: Configure call analysis and scorecards (opens in a new tab)

Deploy

Assign a number to an inbound agent, or import a recipient list into an outbound campaign. WhatsApp conversations are handled in a Twilio-synchronised inbox. SMS goes out from an SMS campaign or a Send SMS step in an automation flow.

Add a number and receive inbound calls

Link a number to its SIP trunk and inbound agent, then place a controlled test call.

Read the guide : Add a number and receive inbound calls
Add a number and receive inbound calls in the dashboard
View full size: Add a number and receive inbound calls (opens in a new tab)

Prepare and import campaign recipients

Upload a CSV with customer_phone and the columns your prompt uses.

Read the guide : Prepare and import campaign recipients
Prepare and import campaign recipients in the dashboard
View full size: Prepare and import campaign recipients (opens in a new tab)

Work with WhatsApp conversations

Find or start a thread in the Twilio-synchronised inbox.

Read the guide : Work with WhatsApp conversations
Work with WhatsApp conversations in the dashboard
View full size: Work with WhatsApp conversations (opens in a new tab)

Manage SMS conversations

Open SMS to review conversations and replies. Use Logs to check message status, Templates for reusable messages, and Settings for sending preferences. Phone numbers are blurred in this screenshot.

Read the guide : Manage SMS conversations
Manage SMS conversations in the dashboard
View full size: Manage SMS conversations (opens in a new tab)

Build and inspect an automation flow

Connect a trigger to steps such as Send SMS, then test before activating.

Read the guide : Build and inspect an automation flow
Build and inspect an automation flow in the dashboard
View full size: Build and inspect an automation flow (opens in a new tab)

Monitor

Every call is recorded in Calls with its transcript and analysis. The dashboard summarises results by date and agent, and its metrics link to the calls behind them.

Use the dashboard to investigate performance

Trace a metric to the call records that explain it.

Read the guide : Use the dashboard to investigate performance
Use the dashboard to investigate performance in the dashboard
View full size: Use the dashboard to investigate performance (opens in a new tab)

Find and export call records

Filter by date, agent, disposition and duration before exporting.

Read the guide : Find and export call records
Find and export call records in the dashboard
View full size: Find and export call records (opens in a new tab)

Review a recording and transcript

Check what was said against the summary and extracted data.

Read the guide : Review a recording and transcript
Review a recording and transcript in the dashboard
View full size: Review a recording and transcript (opens in a new tab)

Request and review a disposition correction

Correct an outcome and keep a reviewed record of the change.

Read the guide : Request and review a disposition correction
Request and review a disposition correction in the dashboard
View full size: Request and review a disposition correction (opens in a new tab)

Developer reference

The developer path is kept separate from the dashboard guides. Create a workspace API key under Configuration → General (Manage workspace preferences and API keys) and send it as a bearer token to the endpoints in the API reference. An agent can also push completed-call data to your own endpoint (Send completed call data to a webhook). To drive the same operations from an AI assistant instead of writing requests, connect the MCP server.

Next steps