Splutter AI

You can change the model behind a live bot without rebuilding it. Splutter connects to OpenAI, Anthropic, and Gemini, and the debugger lets you run the same conversation through two models to compare answers. That's useful when a cheaper model handles routine questions and a stronger one takes the hard ones.

The voicebot runs on speech-to-text, a language model, and text-to-speech, with Deepgram and ElevenLabs handling the audio and OpenAI's Realtime API available for lower latency. You can have the agent call your own phone to test it. Telephony runs through Twilio, so an existing number usually works.

Prompts are split into structured sections instead of one long text box. You define the role, the rules, and the fallback behavior separately, which makes edits safer. Change one rule and you don't accidentally break the rest of the persona.

Every conversation can be inspected turn by turn, with the retrieved context and the model's response side by side. When an answer goes wrong, you see whether the knowledge base was thin or the prompt pushed the model off course. Model comparison lives in the same view.

One omnichannel chatbot serves the website widget, SMS, WhatsApp, Messenger, and Slack. The layout editor handles colors, animation, background, blur, and a custom home screen, so the widget can match your brand rather than looking like a stock chat bubble.

On paid plans you can attach API calls, webhooks, a ticketing tool, a lead capture form, and a booking tool to the bot. A restricted Stripe key plus custom events ties a conversation to a visitor ID, an email, or a phone number, which is how the revenue attribution report gets its numbers.