AI chat AIAI feature

Most of what you do in a day is find something and then change it. The AI chat turns those two steps into two sentences: "find the Mustermann project in Berlin", then "add a task to call them back Tuesday".

The AI chat works on your real data, within your own permissions, and asks before making changes with real consequences.

Before you start

• AI chat is switched on per company. New companies have it from the start. If there's no Reonic AI section in the navigation, ask your Reonic account manager.

Open a chat

Use whichever way suits you:

• From the navigation. Open the Reonic AI section, stay on the Reonic AI tab, and pick a recent chat or click + to start a new one.
• From any Portal page. Press ⌘J (Mac) or Ctrl+J (Windows). A quick-access window headed Reonic AI opens on top of the page you're on, and questions about "this offer" or "this project" refer to that page. Click Open full view (⌘⇧F) to continue in the full chat, or press Esc to close the window.
• From the full chat. Click New chat in the sidebar header.

Type your message in the box at the bottom and press Enter to send. Shift+Enter starts a new line.

A new chat opens on a start screen with suggested prompts. Click one to send it straight away. The start screen also has the attach button and a microphone for dictation.

The sidebar lists your chats with a title, the message count and how recently each chat was active. The title starts as the first words of your first message and is replaced by a short summary once the topic is clear. A Needs input badge marks any chat where the assistant is waiting for your confirmation. Use the search box to filter chats by title, and the trash icon that appears on hover to delete a chat.

Find something

Ask in plain language:

• "Find the project for Mustermann in Berlin"
• "Show me all my open offers from last week"
• "What contacts do we have at Mustermann Solar?"
• "Find any project with a pending grid-registration filing"

The assistant looks in the matching area (projects, contacts, tasks, components, files or activity history) and shows the result as a side panel, a card, a link or plain text. A single project often opens as a project summary in the side panel. Links marked Opens in sidebar open the side panel, and so do Portal links to a residential project, task, contact, wiki page, time-tracking entry, photogrammetry job or catalog component. Other Portal links open the page in the Portal, and outside links open in a new tab.

Narrow the result from there, for example "only the ones in planning", or act on it directly.

Change something

State it plainly. "Create a task on the Mustermann project: call back next Tuesday." "Update the contact at Mustermann Solar: the phone number is now +49 30 12345678."

The assistant finds the record itself, so you don't have to look it up first. If details are missing, the assistant asks for them before going ahead.

Each step shows as a step card, with a spinner while it runs and a check or a warning when it finishes. Consecutive steps collapse into a row you can expand. The assistant confirms in the chat when it's done. If a step shows a warning, read the reply: the assistant usually explains what went wrong, and you can ask again with corrected input.

Confirmation cards

Changes with real consequences pause for your approval. These always ask: moving a project on the kanban, replacing a project's tags, sending an email, sending the customer portal link, completing a checklist segment, and deleting packages or offer variants. Others ask only in the risky case, such as archiving (but not unarchiving), marking a project Lost (but not Won), creating a calendar event that emails attendees, and linking an electricity bill to a project that already has one.

The card is captioned Confirmation required and says what is about to happen and, where it matters, what you lose. Confirm carries out the action and the assistant continues. Decline cancels it, and the assistant acknowledges that and doesn't ask for the same action again. The card then shows a Confirmed or Declined tag.

You don't have to decide straight away. Close the chat and come back later: the card waits for up to 24 hours, and the Needs input badge in the sidebar helps you find it. After 24 hours the card expires and the action doesn't run, so ask again.

Asking for the same thing twice within a minute in the same chat returns the same pending card rather than a second one.

Keep the focus on one project

When you ask about a project or contact by name, the assistant sets it as the working context. A Working in: chip above the input shows its name, for example Working in: Mustermann. From then on you can say "add a task" or "what's the latest activity?" without naming the project each time.

Name another project or contact to switch the focus. To clear it, click × on the chip, or start a New chat. Each chat keeps its own focus.

For a question across projects, such as "how many open offers do I have?", say "across all projects" so the answer isn't narrowed to the project in focus.

The focus never changes what you're allowed to do. Your permissions apply exactly as they do everywhere else. If the assistant can't find the project you named, the focus is cleared rather than left on the previous one.

Dictate instead of typing

1. Click the Record voice microphone in the chat input.
2. Talk.
3. Stop the recording when you're done.
4. Read and edit the transcript in the input box.
5. Send it as usual.

The transcript is added to whatever you've already typed and is never sent automatically. The recording itself is not stored. Only the text you send is kept, like any typed message.

If no speech is detected, you get a notice and can try again closer to the microphone. If your browser blocks microphone access, the button is hidden, or a notice asks you to allow access in your browser settings.

Dictation in the chat is separate from recording a voice note against a project.

Attach a file

The chat accepts images, PDFs, TXT and CSV files. Drag a file into the chat, paste it from your clipboard (handy for screenshots), or click the paperclip (Attach files). The file appears as a chip above the input with a remove button.

Send stays disabled until every upload has finished. A chip with a red "!" failed to upload: remove it and attach the file again. Other file types, such as DOCX or XLSX, are skipped without a chip, so convert them to PDF or CSV first.

Before writing anything from a file, the assistant shows you the data it extracted and waits for your OK. Attached files stay in the chat. To keep one on a project, ask for it: "Save this bill to the Mustermann project's files."

Stop a run

While the assistant is working, Send turns into Stop. Click it, or press Esc twice with an empty draft.

The run stops after any step that is already running. Stop also cancels the confirmation cards still waiting in that chat, so they can't be clicked later out of context. Changes made before you stopped stay in place, so fix those records yourself if needed. You can type a follow-up straight away, and the conversation history is kept.

The assistant also stops itself if a run takes more than 10 minutes or carries out too many actions. The reply then opens with the reason, for example "I stopped this run because it was taking too long." Start again with a more specific request, or split the work into smaller steps.

If the assistant doesn't respond at all, the chat tries once more and then replies "The AI took too long to respond. Please try again with a simpler request."

Things to know

• Check figures, names and summaries before acting on them. An AI can make mistakes reminder sits under the chat input.
• The assistant works within your permissions. Without the deal-value right, the assistant shows no prices or totals and can't edit offers. Without the invoicing right, invoices are left out of answers.
• "How do I…" questions are answered from the Reonic documentation. The assistant repeats click paths exactly as the docs write them and lists the articles it used. A question is never treated as an instruction, so the assistant changes nothing until you ask it to. When the docs don't cover a question, the assistant says so.
• When a request is outside what the chat can do, the assistant tells you. Do that task in the Portal instead.
• Large requests run in batches. A request like "do this to these 40 records" runs in several steps, and the assistant reports how far it got. When a request is too large for one run, the assistant tells you how many records it can handle in the chat and offers to do those. Send the rest as a follow-up request.
• Your chats are yours. Other people in your company can't open them. Your Reonic account manager can read chats in your company.

Last updated on November 9, 2026