Skip to main content
Every Wassist integration is one of three things, depending on who writes the replies and where your code runs:

Webhooks

Your code writes every reply. Wassist forwards each message and you answer through the API.

Managed agents + tools

Wassist’s agent writes the replies and calls your APIs when it needs data or an action.

Apps (alpha)

Your MCP server, packaged so agents can install it, with signed context and API access on every call.

Which one do I want?

  • You already have an agent or bot (LangChain, OpenAI Agents SDK, Vercel eve, a rules engine) → Webhooks.
  • You want an agent live today and only need to plug in your data, like order status or bookings → Managed agent + API tools.
  • You’re building tools that need to know who they’re talking to (loyalty points, reviews, bookings, a helpdesk), which customer, and which conversation → App.
You can mix them. A common setup is a managed agent that answers most questions, an App or API tool for your back office, and a webhook that takes over individual conversations when a human or your own system needs them.
Building a product where your own customers connect their WhatsApp numbers? Both webhooks and managed agents work across many accounts from one organization. See Run WhatsApp for your users.

Webhooks: your code owns the conversation

Set a number’s routing to Webhook and Wassist stops running the agent. Each inbound message arrives at your URL as a signed subscription.message.received event, and you reply with the Messages API.
Routing can be set per number or per conversation. For example, a managed agent can run the number while you conversations.subscribe() a single customer to your webhook.

Hello world

A working handler on the sandbox in 5 minutes.

Conversation routing

Per-number and per-conversation modes, plus lifecycle events.

Webhook reference

Signing, retries, event catalogue and replay.

Managed agents and tools: Wassist owns the conversation

Set routing to Agent and Wassist runs the whole loop: an LLM with your prompt, model and thinking-effort settings, conversation memory and WhatsApp formatting. You extend what it can do from the agent’s Capabilities page: An API tool is just a JSON description of a request:
description inputs are filled in by the model from the conversation, and value inputs are fixed. Values can include %PHONE_NUMBER% and %CALLBACK_URL% (POST to it later to send a follow-up message).

Hello world

An agent replying on the sandbox in 2 minutes.

Configure tools

API tools, connectors, handoffs and more.

Deploy

Put the agent on your own number.

Apps (alpha): installable tools that know the conversation

Apps are in alpha, so the manifest and API may change. Email contact@wassist.app to join the alpha group.
An App is an MCP server plus a manifest (name, icon, MCP URL, who logs in). You install it on your organization and pick which agents get which tools. In the alpha, an App can only be installed on the organization that owns it. Compared with a plain connector, an App adds three things:
  1. Signed context on every tool call. _meta carries the installation, organization, agent and session, plus a JWT signed with your client secret.
  2. API access as the installer. Your client ID and secret, plus the installation ID, give you the whole v1 API for that organization: read the conversation, send messages, look up the contact.
  3. Real auth modes. installer (the business logs in once), end_customer (each WhatsApp customer logs in through a button the agent sends) or none.

API tool, connector or App?

Build an App

Wrap an existing MCP server in 15 minutes, or build a Wassist-aware one.

Example apps

Runnable MCP servers and webhook receivers to copy.