AptiveAI for Apps
AptiveAI for Apps makes AptiveAI the AI provider of an application that has none of its own — your line-of-business system, your customer portal, an internal tool. Instead of wiring a model provider into that application, you configure an agent in AptiveAI Agents (its instructions, its model, its connectors) and let the application use it. Changing a prompt or a model is then a change in AptiveAI, not a release of your application.
It comes in four parts, and you can use any of them:
- The Agents API — your application’s server calls an agent the way it would call a model, through an OpenAI-compatible chat completions endpoint. The agent’s connector tools run on AptiveAI’s side; tools only your application has are handed back to it.
- The embedded chat — a panel of AptiveAI Agents inside your application’s own pages. Your application signs its people in, can tell the agent what the person has open, and can let the agent change what the person sees, such as opening a record.
- The Speech API — live speech-to-text over a WebSocket, for dictation or a voice feature in your application. Your server trades its key for a short-lived token; the browser streams the audio.
- The Provisioning API — your own admin screen manages the AptiveAI account behind your application: adds people (with no password and no email from AptiveAI), sets or frees their seats, removes them, switches services on or off, and checks who could act through each connection to your systems.
The examples on these pages use an invented host application, Northwind Desk, and an invented person, Jane Sample.
Which one do I need
Section titled “Which one do I need”| You want | Use |
|---|---|
| An AI feature inside your application’s own interface — a summary, a draft, a suggestion, an assistant you design yourself | Agents API |
| Your people to chat with your AptiveAI agents without leaving your application, and see their tasks there | Embedded chat |
| An answer computed on your server, with no person present | Agents API |
| An agent that works on a computer with a browser, asks for approvals, or runs for minutes | Embedded chat — a task started there is an ordinary task |
| The agent to open a record or switch a view in your application while it answers | Embedded chat, with screen actions |
| Live dictation or another voice feature | Speech API |
| To add your people to AptiveAI, and set their seats, from your own admin screen | Provisioning API |
The Agents API answers within one request: it has no computer, no approval cards and no stored transcript. The embedded chat is AptiveAI Agents itself, shown inside your application, so everything a task can do, a task started there can do.
Who sets it up
Section titled “Who sets it up”Every part is set up by an owner of your enterprise, in Settings of AptiveAI Agents:
| Section | What it is for |
|---|---|
| Settings → API keys | Keys your application’s server uses, each with the services it may use — Agents API (limited to the agents you choose), Speech API, Provisioning API — and the account’s own switches for the Agents API and Speech |
| Settings → Embedded chat | Turning the panel on, the origins allowed to show it, the signing secret your server signs people in with, and the panel’s colours |
Nobody else sees these sections. The people who use the embedded chat need an AptiveAI account in your enterprise, created by invitation — see Users and roles — or by your server with the Provisioning API.
How it is paid for
Section titled “How it is paid for”The Agents API and the Speech API are paid from your enterprise wallet, per use. Each Agents API request is charged at its usage cost times your enterprise’s rate, and the response says what was charged, in dollars; a speech session is charged by the second, at the prices per hour. Neither ever spends anyone’s seat allowance or promotional credit, and either is refused before it runs when the wallet is empty or almost empty. API calls appear on Activity as the activity API.
A task started in the embedded chat is an ordinary task of the person who started it: it is paid for like any task of theirs — their allowance, then the wallet where the owner allows it — with the same limits and approvals. See Usage and credit. On an account that pays only from its wallet, where nobody holds a seat, the panel’s tasks are paid from the wallet instead, at the Agents API’s rate.
Calling the Provisioning API costs nothing, but assigning a seat when none of that tier is free buys one on the account’s card.
To add funds to the wallet, see Billing and seats.