Chat that knows what it's about.
Every transaction, application, booking or case has two sides that need to talk. ObjectChat gives them a structured conversation, embedded in your product, built around the object that matters, with the actions and guardrails that workflow needs.
Generic chat has no context.
WhatsApp and KakaoTalk have no structure, record or safety guardrails. Intercom and Channel.io are built for one business talking down to many customers, not peer-to-peer, not document-heavy, not state-driven.
Context lives outside the chat
The listing, the booking reference, the case number, they're somewhere else. Agents copy-paste, users repeat themselves, and nothing is connected.
Actions happen outside the chat
A buyer wants to make an offer. A patient needs to upload a document. A visa officer needs to issue a letter. None of it happens in the chat, it happens elsewhere, manually.
Sensitive workflows have no guardrails
Legal, medical, financial, immigration, these need structure, encryption and audit trails. Generic chat provides none of them.
Three concepts. Infinite applications.
Define your object
An object is the thing the conversation is about, a listing, a booking, a ticket, an application. Give it an ID, a type and a state machine; ObjectChat handles the rest.
ObjectChatSDK.mount({ schema })Define your actions
Actions are what can happen at each stage, make an offer, upload a document, issue a certificate. The gateway validates against state. No frontend logic required.
allowedInStates: ['verified']Embed and go
Drop the widget into any page. It reads the schema and renders the right UI, context card, actions, quick replies, state. Your users never leave your platform.
npm install @objectchat/widgetOne widget. Every kind of transaction.
The same infrastructure powers a used-car negotiation, a visa interview, a medical clearance and an HR approval, each with its own object, states and actions.
AI as an action, not a replacement.
A tiered architecture resolves the vast majority of questions before they ever reach a paid model, built as a standalone sidecar, connected as an action rather than baked in.
Four tiers. Most never reach the model.
Keyword resolver
Pattern-matches against your domain, answers ~75% of messages instantly, at zero AI cost.
Semantic cache
Identical or similar questions are answered from cache, no model call needed.
RAG retrieval
Searches your knowledge base and returns answers directly, no generation cost.
LLM synthesis
Only truly novel questions reach the model, less than 5% of all messages.
On top of that: hallucination detection checks every reply against source material and hands sensitive domains to a human when faithfulness fails; request deduplication collapses 50 identical in-flight questions into a single upstream call; and gap analytics surface the questions the knowledge base couldn't answer as draft articles, so the bot gets cheaper and smarter from its own history.
Everything you need. Nothing you don't.
State-driven actions
Validated server-side against object state. "Issue Certificate" only fires when documents are verified; "Mark Sold" only appears when reserved.
End-to-end encryption
For legal, medical, financial and immigration domains, messages are encrypted peer-to-peer. The gateway relays but never reads content.
Real-time with resilience
Binary WebSocket frames, an offline outbox with automatic resync, typing indicators, read receipts and push notifications.
Built-in safety
Scam-pattern detection flags phone numbers and external IDs, verified safe-meetup suggestions, and secure S3-presigned uploads.
Full operator dashboard
Real-time monitoring, agent assignment, canned responses, global search, broadcast messaging and template management.
One embed, any platform
A single JavaScript call, no framework dependencies, persistent anonymous identity, floating launcher or embedded, your choice.
Built for any industry that has transactions.
ObjectChat is domain-agnostic and ships seven pre-built templates to start in minutes. Every configuration is extensible, your industry, your workflow, your rules.
Used Car Sale
Offer modal, inspection request, safe meetup, deal summary
Hotel Booking
Guest-to-staff chat, service scheduling, review prompts
Marketplace
C2C negotiation, meetup coordination, scam detection
SaaS Support
Context-aware tickets, log sharing, escalation, SLA
Visa Support
Document upload, interview scheduling, letter issuance, E2E
Medical Clearance
Records, fitness-to-fly certificates, E2E encrypted
HR Travel Auth
Employee authorization, approval chains, documents
Built to be embedded, extended, and scaled.
A Fastify + WebSocket gateway on DynamoDB, Lambda-deployable via AWS SAM, with a vanilla-TypeScript widget that carries no framework dependencies and a Next.js operator dashboard.