What a Finn is
A Finn is a voice AI agent: a named configuration made of an identity, a voice, a language, a welcome message, instructions (a system prompt or a call flow), knowledge, call settings and post-call analysis questions. A Finn doesn't place or receive calls on its own. It handles live calls only while a deployment is running for it. See concepts.
Create one from the sidebar with Create New Finn. You can start from scratch, from a playbook, or with the AI copilot. See agents-quickstart. Through the API, see api-finns.
Where you configure it
Your Finns are listed under Voice AI Agents in the sidebar. Opening one shows its builder page, with these tabs:
| Tab | What it holds |
|---|---|
| Simulation test | Text conversations with the Finn, saved test scenarios and noise tests. See agents-testing. |
| Call flow | The welcome message, and the system prompt or call flow. |
| Agent handbook | Behavior toggles and the knowledge base. |
| Call setup | Speech and language model settings, and call handling. |
| Post-call analysis | Questions answered from the transcript after each call. |
| Deployment history | Past deployments of this Finn. |
To change the configuration, click Edit on the builder page. The editor has four tabs: Call flow, Agent handbook, Call setup and Post-call analysis.
The configuration blocks
Identity
| Field | What it does |
|---|---|
| Finn agent name | An internal label. Callers never hear it. |
| Voice persona | The voice the Finn speaks with. See agents-voice. |
| Default language | The language the Finn listens and speaks in. See agents-multilingual. |
Welcome message
The first line the Finn speaks when a call connects. See agents-welcome.
System prompt or call flow
The System prompt holds the Finn's objective, flow and rules in plain text. See agents-prompting.
For calls that need fixed steps, or that call tools such as calendar booking, transfer or your own API, convert the Finn to a call flow. When a call flow is active, it runs instead of the system prompt. See agents-tasks.
The welcome message and the system prompt can include {{ variables }} filled from your audience file. Call flow step prompts can't. See agents-variables.
Agent handbook
Toggles that shape how the Finn speaks and behaves, grouped as Personality & Tone, Accuracy & Format and Trust & Safety. For example, Echo verification has the Finn confirm names, numbers and emails, and AI disclosure when asked has it admit it's a virtual agent. See agents-guardrails.
The same tab holds the Knowledge base: files, website pages and pasted text the Finn can draw on. See agents-knowledge-base.
Call setup
| Setting | What it controls |
|---|---|
| Speech to text | The transcription engine. See agents-stt. |
| Language model | The LLM that writes the Finn's replies. See agents-llm. |
| Current time awareness | The time zone the Finn uses for dates and times. |
| Max call duration | A hard cap on call length. |
| Reminder message frequency and Idle reminder message | How long the Finn waits on a silent caller, and what it says to prompt them. |
| End call on silence | Hang up after prolonged silence. |
| Human handoff | A handoff number. Transfers themselves need the Call Transfer tool in a call flow. See tools-transfer. |
See agents-conversation for detail.
Post-call analysis
Questions answered from each call's transcript, such as "Was an appointment booked?". Answers can be Yes/No, a choice from a list, a number or text. See post-call-analysis.
Draft and live
A new Finn is a Draft: it isn't deployed and won't receive live calls. Deploying it makes it live; stopping its deployments returns it to draft. You can test a draft at any time. See agents-testing.
Only a draft Finn can be deleted. A deployed Finn can't be.
Other actions
The More actions menu on the builder page has:
| Action | What it does |
|---|---|
| Duplicate | Copy the Finn into a new one. Use it to try a big change without touching the original. |
| Copy Finn ID | Copy the ID you use in the API. |
| Export config (JSON) | Download the Finn's configuration. |
| Delete | Remove a draft Finn. |
Next
- agents-quickstart: build a Finn and test it.
- agents-prompting: write the system prompt.
- quickstart-inbound or quickstart-outbound: put a Finn on real calls.
- api-finns: the Finn resource over HTTP.