Servly documentation
Servly runs one AI persona across your knowledge, your e-services and the surfaces your customers use. These docs cover how to set it up, embed it and drive it from your own systems.
- QuickstartCreate a knowledge base, upload a document, ask a question and embed the widget.
- InstallationAdd the avatar widget, the browser copilot or the Mendix widget to your app.
- WorkflowsBuild e-services with steps, conditions, validation, OTP and webhooks.
- API referenceEvery endpoint for knowledge bases, documents, Q&A, flows and more.
Core concepts
- Tenant
- Your organization's account. Every knowledge base, document, avatar and flow belongs to one tenant, and every API call is scoped to it by your API key.
- Knowledge base
- A self-contained body of knowledge with its own search index, language, widget settings and embed key. One persona usually maps to one knowledge base.
- Q&A pairs
- The answers the persona gives. They're generated from your documents or written by your team, audited for clarity, and indexed for search.
- Intents
- Responses for the moments that aren't questions: greetings, thanks, goodbyes, out-of-scope requests, “let me check”, and suggestions to start a flow.
- Avatar
- The face and voice of the persona: a portrait, an idle loop and a voice. Answers can be rendered as video clips in that voice.
- Surfaces
- Where people meet the persona: the avatar widget, the browser copilot, and the Mendix widget.
- Flows
- E-services designed in the workflow builder: steps, conditions, API calls and endings, with validation on every input.
- Knowledge gaps
- Questions the persona couldn't answer confidently. They're logged, grouped and ranked so your team can answer each one once.
How a question is answered
When someone asks something on any surface, Servly classifies the intent, searches the knowledge base's index with both semantic and keyword matching, and picks the best answer. If the question matches a flow's trigger, the persona offers to start that service. If nothing matches well enough, the persona says so and the question is logged as a knowledge gap.
Knowledge bases run in one of two modes. Classic mode matches each question to the best prepared answer. Agent mode keeps the conversation's state across turns and uses tools to search answers and flows, ask for clarification, remember details and log gaps.
Base URLs
- API:
https://avatar-api.syntiva.tech - Avatar widget bundles:
https://avatar-cdn.syntiva.tech - Console: console.syntiva.tech