--- title: AI Assistant description: Bob is the in-app AI assistant for ControlCom Connect. Ask it about readings, Alarms, Devices and Assets in plain language, and it answers from live platform data and the documents you have uploaded to your knowledgebase. source: https://documentation.controlcomtech.com/platform/ai-assistant --- # AI Assistant Bob is the assistant built into ControlCom Connect. Ask it a question about your facility in plain language and it answers by calling the platform on your behalf: reading current values, pulling history, checking which Devices are online, summarising Alarms, and searching the documents you have uploaded. > **Workflow Stage: AI** ## Overview The assistant is the platform's AI stage, and it is deliberately the last stage in the workflow: it has nothing useful to say until Devices are Connected, readings are Collected and Stored, and Assets and Locations describe what those readings belong to. Given that groundwork, it turns a question like "are all my devices online?" into an answer, rather than into a trip through four screens. It goes by a few names in the interface. The page header calls it **Bob**, the sidebar calls it **Ask Bob**, the button in the header bar is just **Ask**, and the breadcrumb reads **AI Assistant**. They are all the same assistant and the same conversation. ![The AI assistant page in ControlCom Connect showing the suggested starting questions and the message box](https://documentation.controlcomtech.com/images/small-areas/04-ai-assistant.png) *The assistant on a fresh conversation, with its suggested starting questions* ## Opening the Assistant There are two surfaces. **Ask Bob** at the top of the sidebar opens the full-screen assistant, which is the one to use for a longer piece of work: more room for charts, tool detail and a conversation that runs to several turns. **Ask** in the header bar slides a panel in from the right and leaves the page you were on visible beside it. This is the more common way to use it, because most questions come up while you are already looking at something. The panel is desktop-only and needs a reasonably wide window; it is not available on mobile or in the mobile app. Opening it closes the documentation panel, since both use the right edge of the screen. Both surfaces share one conversation. Ask a question in the panel, then open the full page, and the exchange is still there. Because of that, the panel will not open a second copy of itself on top of the full page: it tells you the assistant is already open and to navigate elsewhere to chat from the panel. ![The Ask panel open on the right of a ControlCom Connect page, with the page still visible beside it](https://documentation.controlcomtech.com/images/small-areas/05-ask-panel.png) *The Ask panel: the assistant alongside the screen you were already working on* The first time anyone in your Organization opens the assistant, they are shown a short disclaimer covering confidential data, third-party processing and the assistant's limits, with a single button at the bottom to acknowledge it. It is per user and shown once. ## Asking a Question On a fresh conversation the assistant offers five starting questions as chips, which double as a fair description of what it is good at: * What is my current kWh reading? * Are there any active alarms? * What was my energy usage last month? * Are all my devices online? * Was there a power outage this week? Below them is the message box, prompting you for what you would like to know or do. There is no model picker, no web search toggle and no file attachment: the assistant reads your platform, not the internet. As it works, the reply is streamed and each step it takes appears as a small card naming the operation and its status, expandable to show the parameters it used and what came back. That is worth reading rather than skipping. A number in an answer is only as good as the point it was read from, and the card tells you which one that was. Two of its steps draw rather than report. One renders a chart of the Variables it looked at directly in the conversation, which you can export as PNG or CSV. The other draws the piece of the model it is talking about as a small tree, switchable between what feeds an Asset, what that Asset feeds, and both. On the last answer, **Retry** asks the question again and **Copy** takes the reply to the clipboard. **New Conversation** on the page, or **New Chat** in the panel, starts over with a clean slate. If an exchange runs long, a banner suggests starting a new one, which is good advice: a long thread costs accuracy, and none of the context you actually need lives in the conversation, it lives in the platform. ## What It Can Reach Everything the assistant says about your facility comes from a call into your Organization's own data, scoped to that Organization. Nothing crosses between Organizations. It can enumerate and search your Assets, read a single Asset in context with its parent, children, Location and typed relationships, and list the Asset Types and relationship types you have defined. It can list Devices with their Variables, search Variables by name, unit, Asset or Device, read the latest value of a point, and pull aggregated history at thirty minute, hourly or daily resolution. It can report which Devices are online, summarise current Alarms and look up Alarm history over a date range. It can list the Locations you have modelled, and fetch current and historical weather for them, which is what lets it put a demand spike next to the temperature that day. It can also list the commands defined on a Device, but that is the extent of it: the assistant is read-only and cannot send a command, change a setting, or create anything. If you want an LLM that can configure the platform, that is a different endpoint with different credentials, documented under the [Builder MCP](https://documentation.controlcomtech.com/build/mcp). ## Grounding the Answers Live data answers "what is it doing". Two settings answer "what does that mean here", and both live in [Settings](https://documentation.controlcomtech.com/platform/settings). **AI Assistant Context**, on the Organization tab, is a free-text description of your facility that the assistant reads at the start of every conversation: the systems you run, the ranges you consider normal, the procedures you follow. It holds up to 32,000 characters and the row shows how much of that you have used. This is the cheapest improvement available to the feature. An assistant told that the essential bus normally sits near 400 kW and that load tests run monthly gives noticeably better answers than one working from readings alone. The **Knowledgebase** tab is the other half. Upload processes, manuals and drawings there, and they are indexed automatically so the assistant can search them and quote the relevant passage back with the document it came from. Drawings are handled as images rather than text, so it reads the actual sheet rather than a description of it. Documents that have passed their review date still surface, but the assistant is required to flag them as possibly out of date, which is why the review cycle on a document is worth setting. Where the two disagree about structure, the model wins. Your Assets, their hierarchy and their relationships are the source of truth for how the facility is wired; the context field is treated as commentary alongside it. ## Limits and Handling of Your Data The disclaimer shown on first use is the short version, and it is accurate. Conversations are processed by a third-party large language model, so your messages leave the platform to be answered. Do not paste passwords, financial records or anything else confidential into the chat. Uploading a document to the Knowledgebase has the same consequence: it becomes material the assistant can send to the model when it is relevant. The assistant can also be wrong. It is instructed to answer only from tool results, to quote units and timestamps, and to check the current state before suggesting any physical action, but none of that is a guarantee. Treat an answer the way you would treat a colleague's recollection: good enough to act on for routine questions, worth confirming on the source screen before anything that matters. ## Availability and Permissions The assistant is enabled per Organization rather than per user, and the switch is not in the app. If your Organization has not been opted in, the page explains that AI features are not currently enabled and asks you to contact your account representative. Once the Organization is enabled, reaching the assistant needs the `AiUsage` permission on your role, which the default administrator and owner roles already carry. See [Team & Roles](https://documentation.controlcomtech.com/platform/team) for how to grant it to others. ## Related Workflow Stages The AI stage sits on top of every stage before it. Devices Connected and Collecting give the assistant points to read, Stored history gives it trends to compare, and the Assets and Locations you modelled are what let it answer about a generator rather than about a topic string. If an answer is thinner than you expected, the gap is usually upstream: an unmodelled Asset, a Variable with logging disabled, or an empty knowledgebase. Alongside it in the Integrate stage, the [Builder MCP](https://documentation.controlcomtech.com/build/mcp) points your own LLM client at the same Organization with tools that can write as well as read, and the [API reference](https://documentation.controlcomtech.com/api-reference) covers direct integration.