> For the complete documentation index, see [llms.txt](https://adatrack.gitbook.io/docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://adatrack.gitbook.io/docs/documentation/features/ai-assistant.md).

# AI Assistant

The **AI Assistant** is a natural-language interface to your AdaTrack platform. Instead of clicking through device lists, telemetry charts, and alert history to answer a question, you can just ask: "which trucks haven't reported in 24 hours?" or "is the temperature in warehouse 3 within normal range?" The assistant has live access to your devices, telemetry, alerts, geofences, and workflows, and answers using your actual data — not a canned response.

It ships as a plugin. An administrator enables it under **Admin → Plugins**; on self-hosted deployments it requires a Standard-tier licence carrying the `plugin:ai_assistant` feature (or `plugin:*`).

***

## Where You'll Find It

Once enabled, the assistant surfaces in four places:

* **Floating chat button** — a sparkles icon in the bottom-right corner of every authenticated page. Click it to open a sliding chat panel without leaving what you're doing.
* **Ask AI buttons** — contextual entry points on the device detail panel, the alert rule builder, the geofence detail page, and the stats page. Clicking one opens the assistant already primed with the entity or page you were looking at, so you don't have to re-explain what you mean by "this device."
* **Dedicated chat page** — open **AI Assistant** in the sidebar for a full-screen view with a session history sidebar, so you can keep multiple conversations going and pick up an old one later.
* **Dashboard widget** — add an **AI Chat** widget to any dashboard for an embedded chat panel, optionally scoped to a fixed set of devices.

Every conversation streams its response token-by-token, and shows what the assistant is doing along the way — "Fetching devices…", "Querying telemetry…" — so you're never staring at a blank panel wondering if it's working.

***

## What It Can Do

The assistant can look up devices and their status, pull recent telemetry readings and the latest reading for a device, list recent alerts, list your geofences, run historical stats queries, trigger a workflow, and summarize overall fleet health (device counts, online status, active alerts). It only ever acts on your own data — every request runs under your account's permissions, the same as if you'd clicked through the UI yourself.

The assistant cannot execute arbitrary code, call arbitrary URLs, or see other users' data. It's scoped to helping you understand and operate your own AdaTrack account.

***

## AI Analysis Workflow Node

Beyond chat, the AI Assistant adds an **AI Analysis** node to [AdaTrack Flow](/docs/documentation/features/workflows.md) (listed under "transform" in the node library). It sends a prompt template through the LLM as part of an automated workflow — for example, summarizing an incident's telemetry history before posting it to Slack. The prompt template supports `{{field}}` placeholders drawn from the workflow's upstream data, and the node's output can be plain text or parsed JSON for downstream nodes to consume.

***

## AI Assistant vs. Chat Bot

AdaTrack also has a separate [Chat Bot plugin](/docs/documentation/features/chatbot.md) for Telegram and Slack, which handles structured commands like `/devices` and `/status`. The two are complementary, not competing:

* **Chat Bot** is command-driven and lives in Telegram/Slack — it's built for quick fleet checks and interactive alert notifications from your team's existing chat tools.
* **AI Assistant** is conversational and lives inside the AdaTrack dashboard (plus the AI Analysis workflow node) — it's built for open-ended questions and multi-step reasoning over your data.

If you enable both plugins and turn on the Chat Bot's **AI-powered replies** setting, free-text messages sent to your Telegram or Slack bot (anything that isn't a `/command`) are routed through the AI Assistant automatically — so your team gets natural-language answers without leaving chat, using the same underlying assistant. If the AI Assistant plugin is disabled, the Chat Bot simply falls back to structured commands only.

***

## Licensing & Usage Limits

Configuring the assistant's LLM provider (Anthropic Claude or OpenAI), API key, and behavior is an admin-only task under **Admin → Plugins → AI Assistant**. Every user with an account can use the assistant once it's configured — it works against their own data only.

Usage is metered against your account's **AI Tokens** quota, the same monthly ceiling shared with the AI Analysis workflow node. See [Quota Management](/docs/documentation/administration/quotas.md) for how the limit is enforced and [Billing & Subscriptions](/docs/documentation/administration/billing.md) for how token usage is billed on metered plans. A short-lived per-minute request limit also applies as a burst-abuse guard, independent of your monthly token budget — if you hit it, the panel shows a brief cooldown before you can send another message.