Dolphy documentation

Dolphy is a customer communication system that searches a business's own knowledge, writes answers grounded in the evidence it finds, and runs permitted business actions through tools. In technical terms, it is a hybrid RAG system with an agent layer for tools and actions.

This does not mean training a new model on customer data. Websites, text and files become a searchable knowledge base. Dolphy retrieves relevant passages, constrains the answer to that evidence and can create a request card or run an approved action.

Where should I start?

• Evaluating Dolphy: Read How it works and Evaluation.
• Setting it up: Follow Quick start, Playground, Knowledge and Channels in order.
• If you will use voice: Voice agent describes what is billable and what you need in place first.
• If you want to see what cannot be changed: Read Rules that cannot be changed.
• For prices: The Pricing page covers the plans and what is actually counted together.

Who is it for?

Dolphy brings two needs into one workspace:

• Teams that answer pre-sales questions and collect appointment or booking requests
• Operations teams handling web, WhatsApp, Instagram and Messenger conversations together
• Managers who need answers grounded in company knowledge and a clear route to a human
• Technical teams connecting read or write operations through webhooks

Infrastructure components

Dolphy uses separate infrastructure components for distinct roles: data and tenant boundaries run on a database stack hosted on our own servers; OpenAI and Google handle language and embedding workloads; voice interaction uses our own live streaming server for call transport, plus separate services for speech-to-text and text-to-speech.

Model and provider versions can change operationally, so the product contract is based on response and safety behaviour rather than a particular model name.