Client project
Calldone: AI voice agents that answer and place business phone calls
Calldone is a voice AI platform and a managed AI receptionist service. Its agents answer business phone calls, book appointments, answer questions and qualify leads, and every call comes back as a recording, a transcript and a summary. It was designed and built by Easital Technologies Ltd. for a client in the United States, which owns and operates it.
The problem
Calls that nobody answers are lost business, and a voice agent that can hold a real phone conversation is hard for a business to build for itself.
Calldone’s About page describes the starting point: businesses lose customers to calls that go unanswered after hours, during rushes or when the team is stretched thin, and voicemail rarely gets a callback.
The second half of the problem is the build. The Calldone home page says: “While other platforms make you build your own AI agent, we handle everything.” The product therefore has two sides: a platform that runs voice agents on real phone lines, and a service in which the Calldone team writes, trains and tests the agent before it takes a live call.
What Easital built
Easital designed and built Calldone for a client in the United States: a platform on which a business can build, deploy and manage AI voice agents that place and receive real phone calls, all from one dashboard.
For Easital, Calldone was a large research and development effort in voice AI. The parts below are the ones the Calldone site describes.
-
Agents
Agents are built with custom prompts, models, transcribers and voices. An agent can be tested live in the browser.
-
Phone numbers
Numbers come from cloud telephony carriers, or from the customer’s own SIP or PBX connection.
-
Tools
During a call an agent can transfer the caller, send an SMS, query a knowledge base, call a webhook or use an integration.
-
Outbound campaigns
Campaigns place outbound calls from a contact list for reminders, follow-ups and surveys.
-
Call review
Every conversation is kept as a recording, a transcript and a summary, with a success evaluation.
How it works
A call runs through speech recognition, a language model and voice synthesis, and ends as a record in the dashboard. Calldone’s terms and privacy policy describe each stage.
| Stage | What happens |
|---|---|
| Call | A caller rings the business’s number and the agent picks up, or a campaign places an outbound call. |
| Listening | Call audio is transcribed by a speech-to-text provider. |
| Conversation | A large language model holds the conversation and generates the content of the reply. |
| Speaking | The agent’s voice is synthesized by a text-to-speech provider. |
| Actions | The agent answers from the knowledge base, books into a calendar, and hands off to a person when a call needs one. |
| Record | The recording, transcript, summary and outcome appear in the dashboard. |
- Languages. The site’s FAQ lists English, Spanish, French, German, Portuguese, Hindi, Arabic and Japanese among the supported languages, and says an agent can be set to detect the language.
- Agent briefs. The privacy policy describes an improvement loop: Calldone periodically takes a sample of an agent’s recent transcripts and asks a language model to write a short brief about how callers to that agent behave and what they ask for.
- Escalation. The customer sets the rules. An agent can transfer the caller to a team member, take a message or schedule a callback.
Technology
Calldone is a real-time voice system that joins telephony, speech-to-text, language models and text-to-speech. Its parts are described here by what they do.
- Speech-to-text: call audio is transcribed during the call, in the languages the agent supports.
- Text-to-speech: the agent’s replies are spoken in a synthesized voice.
- Language models: large language models hold the conversation and write transcripts and summaries. The model, the transcriber and the voice are set per agent.
- Telephony: numbers from cloud telephony carriers, or a SIP or PBX connection the customer configures.
- Integrations: Slack, Google Sheets, Google Calendar, GoHighLevel, Make and custom webhooks.
- Data use: the privacy policy states that call recordings, transcripts, prompts and knowledge base content are not used to train AI models.
Status
Calldone is live at calldone.ai. The site offers a test call with an agent in the browser, with no signup.
Calldone is sold by subscription, with setup and agent training included.
The client owns and operates Calldone. Easital’s part was the design and engineering of the product.
Tell us about the voice AI product you want to build
Send a short description of the product or the problem you want solved. We reply by email with questions and a proposed next step.
Easital is an AI and SaaS engineering company that takes AI software to production, and runs AI products of its own. Founded in 2019.