Using the dashboard

Build your first voice agent

The agent builder end-to-end: write the prompt, pick an engine, voice, and languages, attach tools and knowledge, talk to your draft, and deploy it to production.

Everything about an agent lives in the builder at Voice Agents (/dashboard/agents) — create one and you land in it. This guide walks the same path as the in-app "Build an agent" tour.

1. Write the prompt#

The prompt is the script for your voice agent. Describe — in any language — who the agent is and how it should handle the call: its persona, what it knows, what it should and shouldn't do.

Call actions are prompted too. Transfers, keypad presses, and hang-ups are plain prompt lines like "If the caller asks for a human, transfer the call to +1 555…" — no separate configuration.

2. Pick an engine#

On the Config tab, the engine picker trades off cost and performance across three tiers — Spark is optimized for cost, Bolt for speed, Storm for intelligence on complex prompts. Each card shows its per-minute price. Switch anytime; see the full pricing reference and agent setting reference.

3. Choose a voice and languages#

Pick a voice and play its sample right in the builder. Set the primary language the call starts in, plus any additional languages — the agent switches automatically when a caller uses one. The voice list filters to voices that support every language you've selected. See Supported languages for the full catalog (selected additional languages add 3¢/min).

4. Attach tools and knowledge#

Four sections on the Config tab give the agent capabilities:

• Connected apps — the easiest path: connect Slack, HubSpot, Google & more in a few clicks, then let your agent use them during calls.
• API connections — for any other API, create an API connection for the actions your agent needs (paste a cURL command and the AI wizard drafts the tool).
• Knowledge — attach a knowledge base the agent can search mid-call.
• MCP servers — add an MCP server by URL.

See Connections for how the four kinds differ.

Advanced options below the endpoint sections control supervision, connection tone, tool sounds, who speaks first, voicemail, the call-start consent announcement, and silence check-ins. See Every agent setting.

5. Watch your all-in rate#

The builder shows your all-in per-minute rate — engine plus surcharges — and updates it live as you change settings. No surprises on the bill.

6. Talk to it#

Click Talk and say hello — your agent answers right in the browser (it'll ask for mic access). This is the fastest iteration loop: tweak the prompt, talk, repeat. When you want a caller on the other end instead of yourself, run a simulation.

7. Deploy#

Every edit autosaves to your draft — watch for "Saved" in the top bar. Draft changes aren't live until you click Deploy.

Every deploy is snapshotted in the History tab: inspect what changed between versions and restore any previous one. That makes deploys safe to do often — you can always roll back.

8. Connect it to the world#

The Config tab's Phone numbers and Web widgets sections attach endpoints without leaving the builder — attach an existing number, provision a new demo number, import via VoIP, or link a web widget.

Next steps#