Architecture overview

Chatevo is a multi-tenant SaaS platform for configurable AI assistants. Each organization gets isolated assistants, knowledge bases, tools, and deployments — all orchestrated through a shared pipeline before every reply.

Three-brain orchestration

Every visitor message passes through three specialized LLM stages:

Stage	Role
Intent Analyzer	Classifies what the visitor wants — FAQ lookup, live data, small talk, escalation
Query Strategist	Chooses a path: RAG search, tool call, hybrid (docs + tools), or direct answer
Response Synthesizer	Writes the final reply from retrieved chunks, tool JSON, and your master prompt

Smart query understanding (Starter+) powers the first two stages. See Smart query understanding and Message flow for the full sequence.

Data and retrieval

Document content is chunked, embedded, and stored in Qdrant. At answer time Chatevo runs hybrid search — dense semantic vectors plus keyword matching — merged with Reciprocal Rank Fusion. See Document retrieval architecture.

Uploaded files live in Backblaze B2. Metadata, assistants, deployments, and billing state live in PostgreSQL.

Tools and credentials

API Request and MCP tools call your systems with stored credentials — keys and tokens never reach the browser. Tool results can drive rich cards in the widget via result_transform. See Tools overview and Credentials overview.

Client surfaces

Surface	URL	Purpose
Dashboard	app.chatevo.ai	Configure assistants, KBs, tools, billing
API	api.chatevo.ai	REST API and widget message endpoints
Widget	widget.chatevo.ai/chat.js	Embeddable chat for your site

Replies stream over Server-Sent Events when streaming is enabled. See Widget SSE events.

Tech stack

Layer	Technology
API server	Node.js, Express
Dashboard	Next.js, React
Database	PostgreSQL
Vector store	Qdrant
File storage	Backblaze B2
Billing	Stripe