Four steps from sign-up to a live agent — voice or text — on your site. No engineers, no telephony, no surprise costs.
Four steps. One snippet. Done. — Live in minutes, here's the whole thing.
Name it, give it a role, write its persona, set a greeting if you want it to speak first. Then add your URLs, docs, text, and folders. This is the only thing your agent speaks from, so it stays accurate.
Pick your voice & language.
Preview the voices and choose one. Pick a widget template — or connect your own design via webhook. Set the window title and welcome message. Add tools so it can take action.
Choose a voice and conversation language in AI Settings. Voice & AI are fully included in your plan — no setup, no per-minute metering.
Copy the embed snippet (HTML, iFrame, WordPress, React, or Custom Webhook), drop it on your site, hit deploy. Change anything later and it updates everywhere instantly — no snippet edits.
The widget runs in an isolated shadow DOM so it won't clash with your site's CSS — and you can restrict which pages it shows on.
No servers, no setup, no moving parts to manage. Your agent just runs — reliably — while you focus on your business.
One live voice turn, three stages — and all three are included in one plan.
English streams live, word-by-word captions; other languages transcribe accurately at the end of each sentence.
Transcript + your persona + your knowledge → a short, grounded reply.
Spoken in your chosen voice, streamed sentence-by-sentence so it starts talking before the full reply is written.
Plus automatic turn-taking (detects when the visitor stops) and automatic voice fallback (never goes silent). Per-minute platforms bill STT + LLM + TTS separately.
Key takeaways
An AI voice agent for a website is a chat widget that listens to or reads a visitor's question, generates a grounded reply from your own content, and speaks or types it back — embedded with a single line of code. Speka delivers this as one flat-rate service: it handles speech-to-text, the LLM, and text-to-speech in one plan, answers only from the URLs and documents you provide, and lets you bring your own Gemini API key instead of paying per-minute platform fees.
Paste one <script> snippet into your site's HTML (or use the WordPress, React, or iFrame embed). The agent goes live in minutes — no engineers, telephony, or servers required.
Yes. Speka is fully no-code: copy the embed snippet, drop it on your page, and hit deploy. Edits update everywhere instantly with no snippet changes. See all embed options or read the developer docs.
Add your URLs, docs, text, and folders as the agent's knowledge base. That's the only source it speaks from, which keeps answers accurate and on-brand.
Each turn runs a three-stage Hear → Think → Speak loop: speech-to-text transcribes the visitor, the LLM combines the transcript with your persona and knowledge into a short grounded reply, and text-to-speech speaks it back — streamed sentence-by-sentence with automatic turn-taking and voice fallback. A text chatbot only reads and writes; Speka supports both in a single widget.
Yes. Speka ships embed snippets for HTML, iFrame, WordPress, React, and a Custom Webhook, so it drops into virtually any site or CMS. The widget runs inside an isolated shadow DOM, so it never clashes with your site's CSS, and you can restrict exactly which pages it appears on — useful for ecommerce stores that only want the agent on product or support pages.
| Speka | Typical per-minute voice platform | |
|---|---|---|
| Pricing model | Flat plan, voice + AI included | Separate metering for STT + LLM + TTS |
| Setup | One line of code, no-code | SDK / telephony config |
| Bring your own LLM key | Yes (Gemini) | Rarely |
| Voice and text | Both in one widget | Often separate products |
| Knowledge source | Your URLs/docs only (grounded) | Varies |
| Maintenance | Zero — auto-scaling, 99.9% uptime | Self-managed infra common |
Speka includes speech-to-text, the LLM, and text-to-speech in one flat plan, and you can bring your own Gemini key — unlike platforms that meter each stage by the minute. See transparent pricing with no per-minute fees, the full feature set, or learn more about how it connects to your stack.
Last updated: 5 June 2026
<script> snippet into your site's HTML, or use the WordPress, React, or iFrame embed. The agent is live in minutes — no engineers, telephony, or servers needed. Change anything later and it updates everywhere automatically.