Documentation / Voice agents
Voice agents
Build, connect and monitor voice agents with Tigy AI.
Currently, available agents are configured through a prompt. Define behavior in the agent instructions and add tools and documents in the settings.
Voice agents accomplish tasks through conversation: answering questions, retrieving information and taking action in your systems. In Tigy AI, you define instructions, choose voices and models, connect tools and review agent runs in one workspace.
Platform capabilities
Explore the guides for each stage, from configuration to operating your agent.
Build and configure
| Goal | Guide | What you configure |
|---|---|---|
| Write instructions | Prompting | Define goals, tools and boundaries with examples. |
| Tune turn-taking | Conversation | Review silence, interruptions and duration. |
| Test before publishing | Testing | Verify responses using voice and text. |
| Build your first agent | First agent | Define a task, write instructions and test in your browser. |
| Define agent behavior | Instructions | Set what to ask, how to respond and when to take action. |
| Choose a language model | LLM | Configure the model that interprets the conversation and generates responses. |
| Choose a voice | Voice | Define how the agent speaks to users. |
| Configure transcription | Transcription | Choose how user speech is converted to text. |
| Add documents | Knowledge base | Provide information the agent can retrieve. |
| Connect actions | Tools | Enable API lookups and actions during the conversation. |
Connect and deploy
| Goal | Guide | How to connect |
|---|---|---|
| Publish changes | Publishing | Validate a draft and publish a version. |
| Review changes | Versions | Inspect history and compare configurations. |
| Receive phone calls | Telephony | Configure a provider and connect your phone line to the agent. |
| Route incoming calls | Inbound calls | Set up handling for incoming calls. |
| Integrate an external API | HTTP API | Define endpoints and parameters for tools. |
| Transfer a call | Call transfer | Route the call when another handler is needed. |
| Send data after a conversation | Webhooks | Deliver call data to your systems. |
Monitor and improve
| Goal | Guide | What to review |
|---|---|---|
| Inspect conversations | Runs and usage | Review status, transcripts, duration and recordings when available. |
| Evaluate quality | Runs and usage | Review transcripts and tool calls against your goal. |
| Investigate a failure | Troubleshooting | Identify the component and verify a fix. |
| Track results | Reports | Review reports available in your workspace. |
| Understand consumption | Billing | Track credits, overages and subscriptions. |
| Refine responses | Test and refine | Test questions, review responses and improve instructions. |
How conversations work
During a conversation, three components work with the agent's instructions and context:
- Speech to text (STT): converts user speech into text.
- Language model (LLM): interprets the request and prepares a response; it can retrieve documents or call enabled tools.
- Text to speech (TTS): converts the response into audio.
Instructions define how the agent should act. The knowledge base supplies information and tools connect conversations to your systems.
Start here
Follow Your first agent in 5 minutes to build an agent and test a browser conversation. Then add tools and documents or connect telephony.