Skip to content
Falcon Nest

Your first production voice agent, live today.

Falcon Voice Core

A complete real-time voice agent for a single conversation at a time. Bring your own telephony or use the browser widget, point it at your knowledge base, and take live calls the same day.

  • Sub-400 ms typical response latency
  • 6 languages with automatic detection
  • 1 concurrent conversation
  • Unlimited transcripts and exports
Choose a plan

Licence keys are issued the moment payment clears. Prices exclude VAT, which is calculated at checkout from your billing country.

Overview

Falcon Voice Core is the shortest path from an idea to a voice agent that real customers can talk to. It bundles speech recognition, turn-taking, reasoning and speech synthesis into one runtime, so you are not stitching four vendors together and debugging latency between them.

The agent listens continuously, detects when the caller has finished speaking, and responds in under 400 milliseconds end to end on a warm connection. Interruptions are handled the way people expect: if the caller talks over the agent, the agent stops, listens, and picks up the thread.

Core is deliberately scoped to one concurrent conversation. That is enough to validate a use case, run a pilot on a single support line, or embed a voice assistant in a product demo. When you need parallel calls, phone number provisioning and analytics, Falcon Voice Pro is a one-click upgrade that keeps your prompts, tools and transcripts.

  • Deploy in the browser with a two-line embed, or connect your own SIP trunk
  • Define behaviour in plain language, then attach tools for the actions that matter
  • Every call is transcribed, timestamped and exportable

Capabilities

  • Genuinely real-time

    Streaming recognition and synthesis run in parallel rather than in sequence, keeping typical end-to-end response under 400 ms so conversations do not feel like radio calls.

  • Natural turn-taking

    Endpoint detection distinguishes a pause for thought from the end of a sentence, and barge-in lets callers interrupt without the agent talking over them.

  • Grounded in your content

    Point the agent at documents, URLs or a help centre export. Answers cite the source passage so you can see exactly why it said what it said.

  • Tool calling

    Give the agent typed functions to look up an order, check availability or create a ticket. Arguments are schema-validated before your endpoint is called.

  • Full transcripts

    Every conversation is stored with speaker labels, timings, tool calls and the model's reasoning summary. Export as JSON or CSV at any time.

  • Browser or telephony

    Embed the WebRTC widget in any page, or bring your own SIP trunk. The agent behaves identically on both transports.

Where it earns its keep

  • Pilot a support line

    Put a voice agent in front of your first-line queue for one product area and measure containment before you commit to a wider rollout.

  • Appointment booking

    Handle scheduling calls end to end: check the calendar tool, confirm a slot, and read back the details for confirmation.

  • Talkable product demos

    Embed a voice agent in your marketing site so prospects can ask questions out loud instead of reading documentation.

Questions

How quickly can I have an agent taking calls?
Most teams have a working agent in under an hour. You describe the behaviour in plain language, upload or link your reference content, and embed the widget. Connecting a SIP trunk usually adds a day depending on your carrier.
What happens if I exceed 1,500 minutes in a month?
Calls keep working. Overage is billed against a Voice Minutes Pack, which you can add at any time. We email you at 80% and 100% of the included allowance so nothing is a surprise.
Can the agent transfer to a human?
Yes. Warm transfer is a built-in tool: the agent can hand the call to a number or SIP address and pass a short summary of the conversation so far.
Where is call audio processed and stored?
Audio is processed in the region you select at setup (UK, EU or US) and is not retained after the call unless you explicitly enable recording. Transcripts are retained for 90 days on this tier.
Is Core enough for production, or only for testing?
It is a production licence. The only hard limit is one conversation at a time, which is genuinely sufficient for low-volume lines. If two callers arrive at once, the second hears your configured busy behaviour.