Skip to main content
Connic
Back to BlogAnnouncement

Introducing Connic Voice: Managed AI voice infrastructure

Build voice agents with your knowledge and tools, choose a realtime model, and use Connic connectors for incoming calls on managed voice infrastructure.

September 19, 20267 min readAuthor: Connic Engineering

Connic Voice is managed AI voice infrastructure for building phone agents with your prompts, knowledge, and tools. Choose a supported realtime model, connect incoming calls through Connic connectors, and let Connic run the infrastructure.

Native speech-to-speech, built into Connic

A caller explains a problem, the agent asks a follow-up question, and the conversation continues out loud. Native speech-to-speech models receive audio and produce spoken replies. The agent can also look up information or use a tool while helping the caller.

Connic hosts the voice agent, connects incoming calls, and handles the live audio connection. Teams configure the agent in a Connic project, deploy it, and inspect its conversations in the dashboard. There is no separate voice infrastructure for the team to operate.

A support agent can use company product knowledge and connected business services during a call. Its prompt defines its role; the connected knowledge and tools determine what it can help with.

Choose your model and provider

Connic Voice supports OpenAI Realtime and Gemini Live, with provider connections through OpenAI, Azure OpenAI, Google Gemini, and Vertex AI. Teams choose a supported native audio model and supply their own provider credentials. Available models and voices depend on the provider and the account’s access.

The new support template starts with Gemini 3.8 Live. Its model selection, prompt, knowledge, and tools can all be changed as the use case develops. Switching to another supported provider also means checking that provider’s voice options and transcription settings. The voice documentation lists the supported configuration.

Connect your knowledge base and business tools

A useful support call needs specific answers: where to find an invoice, which recovery steps to try, or what a cancellation policy allows. A voice agent can search a company knowledge base during the conversation and use the returned information to answer. With Connic Retrieval, those answers can draw on the documents available to the agent.

Tools let the agent act on a request. A team can connect a Python function to its helpdesk API or give the agent access to an existing MCP tool. For example, a configured support integration could look up an order or create a ticket during the call. The team supplies that integration and decides which actions the agent may take.

An agent that explains a refund policy needs access to the policy. An agent that issues refunds also needs an authorized tool for that action. The starter below deliberately uses a mock ticket tool so the first test calls can explore the conversation without creating helpdesk requests.

Shape the conversation

A phone conversation needs a different pace from a written answer. The support template’s prompt asks the agent to give short replies, ask one question at a time, and check whether a troubleshooting step helped before continuing. Teams can change those instructions to match their service and tone.

Voice settings control how the call feels:

  • Voice and greeting: choose a provider-supported voice, set a preferred language, and write the opening message.
  • Interruptions and pauses: let callers interrupt a spoken reply and adjust how long a pause counts as the end of their turn.
  • Waiting for an answer: keep a quiet thinking sound enabled during tool waits or model-reported background work. It pauses while the caller or agent speaks.
  • Ending the call: set an idle timeout and choose whether the agent may hang up. When it ends a call, Connic lets the goodbye finish playing first.

Connect incoming calls with Connic connectors

A Connic connector that supports voice links incoming phone calls to a deployed agent. Teams configure their phone service according to the connector’s setup instructions. Each voice connector targets one deployed voice agent; Connic handles the live conversation.

Build a customer support voice agent

The new Voice Customer Support template searches your support documentation with Connic Retrieval and creates mock support tickets. Add your help articles and policies to Retrieval, and the agent can use them during a call.

Try a support conversation

Add your account-recovery guide to Connic Retrieval, then start a test call with “My password reset link has expired.” The prompt asks the agent to search Retrieval and explain the recovery steps from your guide.

Then try “Those steps didn’t work. Can you create a demo ticket?” The agent is instructed to collect a fictional email address and issue summary, read the details back, and ask for confirmation before using the mock ticket tool.

The tool returns a simulated reference beginning with DEMO. It does not create a helpdesk ticket, notify a support team, or transfer the caller to a person. Use fictional details, since tool inputs and results can appear in conversation traces.

Customize the prompt and greeting for your service, then connect the ticket tool to your helpdesk. Manage the support knowledge in Connic Retrieval: upload documents, paste text, or connect a source that syncs your help content. Knowledge updates become available after indexing, without redeploying the agent.

Inspect transcripts, tool calls, and usage

After a call, open its run in the dashboard to review the conversation and follow the agent’s actions. See which knowledge the agent retrieved, which tools it used, and what each tool returned. The runs and traces view gives teams the detail they need to refine answers and improve the support experience.

Gemini and Vertex AI provide native transcription. For OpenAI and Azure OpenAI, teams configure a transcription model to capture the caller’s words. Audio usage details reported by the provider help teams understand consumption. Reviewing the trace alongside a test call helps fine-tune both the answers and the conversation’s pace.

Deploy the example and make a test call

  1. Install the support template. Create a local project from the example with the Connic Composer SDK:

    pip install --upgrade connic-composer-sdk
    connic init my-voice-support --templates=voice-customer-support
    cd my-voice-support
    connic lint
  2. Save model credentials before deploying. In the dashboard, create a Connic project and open Project Settings → LLM Provider. Save Google Gemini credentials with access to the template’s model, or configure a different supported provider and model.

  3. Add your support knowledge. Select the environment you will deploy to and open Retrieval. Add your support documents with the namespace support, or sync a source into that namespace. Wait for indexing to finish, then use Search to check a question your documents answer.

  4. Deploy and connect a number. Follow the template’s Git or CLI deployment steps, then add a Connic connector that supports voice and link it to the deployed agent.

  5. Call and review. Ask a question covered by your support documents and try a mock ticket request using fictional details. Open the resulting run to review the Retrieval results and ticket receipt, then adjust the prompt or knowledge and test again.

Build your first Connic voice agent

Start with a customer support agent, make a test call, and adapt its knowledge, tools, and conversation to your service.

Get the voice support template

More from the Blog

Tutorial

Expose the Invoice Processor Template as an MCP Tool

Deploy Connic's invoice template as an MCP tool, discover it from Python, send invoice text, and check the extracted fields with a complete client example.

September 11, 202610 min read
Announcement

Introducing Connic MCP: Operate Connic from a Coding Agent

An MCP coding agent such as Codex, Claude Code, or Cursor can connect to a Connic project for debugging runs, shipping deployments, and managing agents without leaving the editor.

August 10, 20267 min read
Announcement

Introducing Connic Models: Every Model EU-Hosted, Zero Config

Connic Models put 30 EU-hosted models from nine creators behind connic/* IDs. Reference one in agent YAML without a provider account, key, or region flag.

July 30, 202610 min read
Industry Insights

The OpenAI Hugging Face Hack: Guardrail Lessons for AI Agents

OpenAI models escaped a test sandbox and breached Hugging Face in July 2026. What the incident reveals about guardrails and how to secure production AI agents.

July 24, 20269 min read
Announcement

Introducing the Connic Marketplace

The Connic Marketplace is the new home for agent templates, connectors, and retrieval sources. One catalog, one publisher model, one CLI install.

July 5, 20265 min read
Industry Insights

EU AI Act Enforcement: Who Investigates and What Evidence to Keep

The AI Office, national authorities, and EDPS divide EU AI Act enforcement by system and provider; teams should keep scoped governance and runtime evidence.

April 13, 202614 min read
Changelog

What We Shipped in March 2026

A/B testing, agent guardrails, API spec tools, dashboard templates with percentile metrics, migration CLI, and more.

April 1, 20266 min read
Changelog

What We Shipped in February 2026

Managed database, templates library, evaluation judges, Telegram connector, web page reading, persistent sessions, conditional tools, and concurrency rules.

March 1, 20266 min read
Changelog

What We Shipped in October 2025

10 deployment regions across 5 continents, import predefined tools (trigger_agent, retrieval_query, web_search) in custom Python code.

November 3, 20254 min read