WhatsApp AI agents that resolve, not just reply
Connect AI agents are conversational AI workers for WhatsApp, built on LangChain and LangGraph with persistent memory. They understand intent, answer from your own knowledge bases (RAG), call your APIs and tools, create tickets, notes, and reminders, and book appointments, then hand off to a human with full context. Model-agnostic across OpenAI, Anthropic, and Google Gemini.
What this changes for your team
- Resolve instead of deflect: fewer stuck customers, fewer escalations.
- Answer from your truth: RAG knowledge bases keep responses accurate and on-brand.
- Act, don't just inform: tools let the agent do the task, not describe it.
- No model lock-in: choose the best LLM per task and swap freely.
From setup to running
- 01
Define the agent
Write a system prompt, pick a model and parameters, and attach a knowledge base and tools in the builder.
- 02
Give it knowledge
Connect knowledge bases so the agent retrieves answers from your documents (RAG) instead of guessing.
- 03
Give it tools
Attach HTTP/API tools and built-in actions (tickets, notes, reminders); an orchestrator routes to the right sub-agent.
- 04
Test, then deploy
Validate in a live playground with token/cost visibility, then invoke inline, in a workflow, or as the always-on responder.
What's inside
Agent types
Conversational responder, summarizer, evaluator, ticket/note/reminder creators, text enhancer, and fully custom agents.
Persistent memory
LangGraph checkpointer memory keeps multi-turn context across a conversation.
RAG knowledge bases
Ground answers in your own content, with retrieval tuned by the number of results returned.
Generic tools framework
Outbound HTTP/API tools, RAG retrieval, and an orchestrator, validated against JSON schemas, so agents act in your stack.
Interactive replies
Agents return ready-made WhatsApp interactive messages (buttons and lists), not just plain text.
Multi-LLM, model-agnostic
OpenAI, Anthropic, and Google Gemini through one framework, with per-model cost and token accounting.
Test sessions + cost tracking
A playground accumulates input/output tokens and computes cost per model before you go live.
Clone to start fast
Clone a public default agent and customize it for your business.
Where it fits
Common questions
How is a Connect AI agent different from a rules chatbot?
A rules chatbot follows a fixed decision tree and deflects when a customer goes off-script. A Connect agent reasons over the conversation with memory, retrieves answers from your knowledge bases so it stays accurate, and calls your tools to take action, then hands off to a human with full context when appropriate.
Will the agent make things up?
The agent answers from the knowledge bases you attach (retrieval-augmented generation), which keeps responses grounded in your content rather than guessing. You also control it with deterministic flow steps around the agent and human handoff.
Can the agent actually do things, or just talk?
It acts. Through the tools framework it calls your HTTP APIs, creates tickets, notes, and reminders, retrieves from knowledge bases, and books, reschedules, or cancels appointments.
Which AI models can I use?
Connect is model-agnostic: OpenAI, Anthropic, and Google Gemini through one framework. Choose the best model per task and switch without rebuilding the agent.
Can I test an agent before customers see it?
Yes. The test playground runs a live chat and shows token usage and cost so you can validate behavior and edge cases before deploying.
How do I put an agent into a conversation?
Three ways: mention it inline from the shared inbox with @, add it as a node inside a workflow, or set it as the always-on responder for a WhatsApp number.
Ready to put an AI workforce on your frontline?
Deploy voice, chat, and video agents that work 24/7, act in your systems, and hand off to your team when it counts. Build your first agent in minutes.
White-glove onboarding for teams · Talk to a human, not a form.