Skip to content

In-app AI assistant Help users who get stuck inside your app.

An assistant inside your signed-in app that knows who each user is. It answers from your docs and does things for them through your own API, such as creating a project, only as that user.

Best forSaaS products that want an assistant inside their signed-in app, acting for each user.

Start free →How it's built

ExampleInside Beacon's signed-in app, the assistant knows who the user is: it answers from the docs, and creates projects for them through Beacon's own API, as that user and no further.

STARTvisitor messageCConditionintentcreateotherAtAsk the Visitorproject_nameAtAsk the VisitorwebsiteTCTool Callcreate_projectCMChat Modelcreated_reply

Each step lights as the agent runs it.

BBeacon AssistantSigned in as Leo
Ask a question…↑
How it's built

What it knows, what it can do, and the steps it follows.

The same three parts make every chatbot, set up in the dashboard without code: knowledge you add, tools you switch on or connect, and a customer flow you draw in the visual editor.

01

Knowledge

How the product works comes from the docs, with the source. What the user has comes from your API.

  • docs.beacon.appWebsite · 60 pages
  • docs/sharing.mdMarkdown · 7 passages
  • Campaigns guide.pdfPDF · 24 passages
02

Tools

The widget goes in your app with one script tag, and passes the signed-in user's token on to your own server.

Your APIYour own endpoints as actions: the chatbot calls them with your keys, which stay on our server and never reach the model.create_projectlist_projects
Your tool serverOr the tools your MCP server offers. Your site signs a token for its user, and only your server sees it.create_project
03

Workflow

The customer flow, drawn in the visual editor: a condition reads what the visitor wants, and each scenario has its own path. A question none of them takes is answered from the knowledge.

Before you set it up

What people ask about an in-app assistant.

What does an in-app assistant cost?

You can try it on the free Hobby plan, with 50 answer credits a month and no card. Pro, at $29 a month, gives 7,000 credits, about 7,000 answers on the default model, GPT-6 Luna, and Business, at $49, gives 20,000. An answer that calls a tool uses a little more, since the model reads what the tool returned, and a long conversation uses more per answer.

What do I need to set it up?

The widget's script tag in your app, and the actions it may take: your own API endpoints, or the tools of a tool server (MCP) you run. Your site signs a token for the signed-in user, which the widget passes on to your server, so every call acts as that user.

What can't it do?

Only what your actions or tool server allow, and only for the user whose token it carries. Your API keys stay on ChatFrom's server and never reach the model.

Does it work in a mobile app?

Yes. The ChatFrom iOS SDK puts the same assistant in your iPhone and iPad app, as a bubble or a row in a list, and passes on your signed-in user's token the same way. With RevenueCat connected, a customer can also ask about their own subscription there.

More use cases
Customer supportOrders and returnsSubscriptionsBookingsLeads and quotesInternal assistantIt can be anything →