THE PLATFORM

A conversational AI engine built around context - not just a language model with a transcript bolted on.

Architecture that takes context seriously

Most conversational AI APIs accept a message array and return a response. That's not dialogue management - it's stateless text completion. Equmenopolis maintains a typed session state that persists entities, tracks resolved intents, and carries coreference chains across the full conversation.

When your user says "fix that third issue we discussed," the platform knows what "that" refers to, which issue index "third" maps to, and that "we discussed" is a past-session reference. It doesn't hallucinate a plausible answer - it resolves it from the tracked state.

Equmenopolis Platform Dashboard

CORE COMPONENTS

01

Intent Recognition Engine

Maps utterances to structured intents with entity slots. Handles multi-intent messages: "book a flight and cancel my hotel" produces two separate intent objects with shared context. Intent confidence scores are exposed in the API response for downstream routing decisions.

  • 300+ pre-built intents across support, commerce, scheduling, and knowledge domains
  • Fine-tunable on your domain with as few as 50 labeled examples per intent
  • Intent correction: recognizes when users contradict a previous intent
02

Context Memory Store

Session state is a first-class typed object, not a raw string. Entities are tagged with type, value, and the turn they were introduced. The memory store supports explicit lookups: retrieve "all entities of type PRODUCT mentioned in turns 1-5."

  • Entity types: PERSON, PRODUCT, DATE, ORDER, ISSUE, LOCATION, and custom types
  • Coreference resolution: pronouns and definite references resolve to tracked entities
  • Session persistence: context survives page reloads and re-engagement after 24 hours
03

Multi-Turn Dialog Manager

A finite-state dialog manager handles conversation flows that span multiple turns. Define flows as JSON state machines: states, transitions, slot-filling requirements, and fallback handlers. The manager runs server-side and drives the response generator.

  • Slot filling: prompts for missing required information before advancing the flow
  • Clarification handling: detects ambiguous inputs and generates targeted clarifying questions
  • Topic switching: handles mid-flow topic changes and returns to the original flow after resolution
04

REST API

Single POST endpoint. Send a session ID and user message, receive a structured response with intent, entities, dialog state, and generated text. No streaming required for standard turns. WebSocket available for real-time applications.

  • P95 latency: under 80ms for standard turn processing
  • Response includes: intent, confidence, entities, context diff, generated text
  • SDKs for Python, Node.js, and Go. REST works with any HTTP client

INTEGRATIONS

Customer Support Platforms

Native connectors for Zendesk, Intercom, and Freshdesk. Dialog state syncs to ticket context. Handoffs to human agents include the full entity and intent trace from the AI session.

Chat Surfaces

Drop-in widget for web apps. SDK handles session management, message routing, and UI rendering. Override the default UI or use the API directly for custom front-ends.

Data Pipelines

Export turn-level analytics to your data warehouse. Schemas for BigQuery, Snowflake, and Redshift. Every turn, intent, entity, and context operation is logged and queryable.

Webhook Events

Fire webhooks on dialog events: intent matched, slot filled, flow completed, fallback triggered. Use these to drive backend actions without polling the API.

Read the API docs

Full reference for the dialog API, context store, and webhook events.

Start Free - 10,000 turns