Build & deploy
voice, video & text agents
in minutes, not days
Let us worry about engineering, reliability & scalability, so that you can focus on building quality AI agents.
No code, no infrastructure
Describe the agent and publish it. We compile that definition and run the realtime session behind it.
Bring your own keys
Your provider accounts bill you directly, at your rates. Our platform fee stays small enough never to eat your gross margin.
Focus on what matters
Prompts, tools, MCP servers, triggers and models — the decisions that actually decide whether an agent is any good.
Your own keys for
OpenAI
Gemini
xAI
OpenRouter
Deepgram
ElevenLabs
Soniox
Sarvam
Raya
Anam
ai-coustics
Twilio
Plivo
Exotel
VobizChannels
One definition. Three ways to talk.
The prompt and the tools do not change when the channel does. Switching it changes which model slots the agent fills, not how it behaves.
Text
Chat on WhatsApp, on Telegram, in a widget embedded in your app or straight over the API, with the same tools behind it.
- WhatsApp numbers and Telegram bots
- Streamed replies over SSE
- Images in, grounded answers out
Voice
A browser call, a real phone number or a WhatsApp call, running either a speech pipeline you assemble yourself or a single realtime model.
- STT + LLM + TTS, or one realtime model
- Inbound and outbound PSTN on your carrier
- Recording and post-call analysis
Video
The same voice agent wearing an Anam avatar, so the person on the other end has a face to talk to.
- Anam avatars, picked per agent
- Identical prompt and tools
- Browser sessions, no plugin
Capabilities
Everything you would have written by hand.
Build it yourself in the dashboard, hand it to Copilot and review what changed, or drive the whole thing from Claude Code over MCP. Same agent either way.
Model stack
System prompt
You handle refund requests for Northwind. Confirm the order number before you look anything up, and never promise a refund the policy does not allow. You are speaking with {{userdata.name}}.
Tools
- look_up_orderHTTP
- issue_refundCode
- transfer_to_humanTransfer
MCP servers
Calendly
HubSpot
FAQs
Tools that do real work
HTTP requests, code steps, conditions, generated replies, frontend actions, MCP operations, handoffs and end-call — composed as a tree, not a script.
Phone numbers, your carrier
Connect Twilio, Plivo, Exotel or Vobiz, import your DIDs, and point inbound calls at a published agent. PSTN billing stays with the carrier.
Copilot in every editor
An AI builder that can rewrite the prompt, swap models, add tools and publish — and shows you exactly what it changed.
Answers in your words
Write the questions people ask and the answers you want given. The agent looks them up mid-conversation and says them the same way every time.
Sessions you can audit
Transcript, tool invocations, final userdata, per-provider usage and a cost breakdown for every session — test runs included. Signed webhooks push the same events to your own systems.
API and MCP, not just a UI
Every dashboard action is an API operation, and the same set is exposed over MCP — so Claude Code or Codex can build the agent with you.
Platform fee
One flat fee. Nothing else.
Talqing is strict bring-your-own-keys, so model, speech, avatar and carrier costs go straight to your own accounts at your own rates. A platform fee — per minute of a call, per message of a chat — is the only thing we add on top, and the only thing we ever charge you for.
Text
$0.0001per message
Charged for each message your agent answers, on the API, WhatsApp or Telegram. A chat left open costs nothing while nobody is writing.
Voice
$0.0035per minute
Browser, phone and WhatsApp calls alike. Your carrier bills the PSTN leg; we bill the agent that ran on it.
Video
$0.01per minute
The same voice agent with an Anam avatar attached, billed for the minutes the avatar is on screen.
Discounts are available at committed volume on both voice and video.
Talk to us about volumeFAQ
Questions we get
before the first call.
What can I build with Talqing?
Voice, video and text agents that talk to your users — on a phone number, in the browser, on WhatsApp or Telegram — call tools, fire webhooks, capture session state and hand off to other agents. Realtime voice is the centre of gravity; the same agent model also runs unattended tasks that take typed inputs and return a typed result.
Do I need to write LiveKit Agents code?
No. You author the agent in the dashboard, through Copilot, or over MCP from your own editor. Talqing compiles that definition into a LiveKit AgentSession and runs it on the platform runtime.
Whose API keys does it use?
Yours. Talqing is strict bring-your-own-keys: you connect your own OpenAI, Gemini, xAI, OpenRouter, Deepgram, ElevenLabs, Soniox, Sarvam, Raya and Anam accounts, and those providers bill you directly. We never resell tokens or minutes — the platform fee is the whole of what we charge.
Can I test an agent before publishing?
Yes. Run browser voice, video and text sessions from the dashboard, then read the transcript, tool calls, captured fields, provider usage and cost for that session before you publish anything.
How do tools work without code?
A tool is an operation tree. You compose HTTP requests, conditional branches, code steps, generated replies, frontend actions, MCP and provider operations, handoffs, end-call and webhook triggers from the editor — and there is a code operation for the cases that genuinely need one.
Is telephony included?
Yes, on your own carrier account — Twilio, Plivo, Exotel or Vobiz. Connect the account under Phone Numbers, import your DIDs, assign a published voice agent for inbound, and place outbound calls one at a time or as a scheduled batch, from the dashboard or the API. If a contact-centre platform already holds your numbers, it can stream call audio to the agent over a WebSocket instead. PSTN charges stay with your carrier; Talqing runs the agent.
Your first agent
is one prompt away.
Sign in with Google, tell it what the job is, and place a test call before your coffee goes cold.