Analytics Agent integrates closely with Xedant Agent — the shared AI assistant of all your applications. The agent chat and the unified sidebar are available on every page of Analytics Agent: reports, charts and tables are added in plain language, errors are fixed with the “Fix with AI” button, and chains of tasks can be put into a queue. The integration is optional: without it, everything else works.
Enabling
Set the AGENT_API_URL and AGENT_API_KEY variables — the URL and API key of your installed Xedant Agent. Without them the AI buttons are hidden; the rest of the functionality doesn’t suffer.
One assistant — several applications. Analytics Agent uses the Xedant Agent you already have configured: the same models, keys and skills as in all your other applications. No separate LLM subscriptions or keys for analytics — everything is configured once, in the center.
Adding things in plain language
- describe in words what you need: “add a weekly sales chart”, “break the table down by region”;
- an agent chat opens with a prepared prompt: the report URL, the document path, the block ID;
- pick an AI model from the list — and send the request.
How the agent edits a report
The agent edits the report’s JSON directly — through the same mechanism as a manual edit. The file watcher picks up the file change, and the block appears on screen instantly, without reloading the page. The “ask → see” cycle takes seconds.
Fix with AI
Next to validation and block errors there is a “Fix with AI” button: the error text is placed into the prompt automatically, the agent receives the block’s context and fixes the JSON. The mechanism is the same for all buttons — fixWithAI.
AI models
- the model list comes from your Xedant Agent — the same models and settings as in all the other applications;
- the model is picked when sending a request;
- managing models and their colors is done in the “Models” section of Xedant Agent itself.
The sidebar and chats on every page
Analytics Agent has one sidebar visible on every page: the section tree (Documents, Templates, Data Sources, Lakehouse, Alerts), document search and recent history. The list of active chats and the “Prompts” button live here too. The agent chat opens from anywhere — no switching between applications.
- from a report you can start a chat about a specific document in one action — the system fills in the file path and the page address for you;
- from any analysis template — press “Use this template”: a chat opens with a ready request to apply the template to your data (details in Analysis Templates);
- the agent’s progress is visible in real time: messages, tools, processing time; sound notifications about new messages are available (the sound is turned off with one button);
- chats can be tagged with colored labels: yours, important, in progress — and the list filtered by them; labels sync with Xedant Agent;
- for every request you pick the AI model and the skill from your Xedant Agent’s list;
- the full chat history is saved and available from the same sidebar.
The prompts queue and auto-send
“Prompts” is a list of tasks the agent will complete one by one. You write the tasks in plain language, turn on “Auto-send” — and the agent takes the tasks one after another until the list runs out. For example, in the morning you can drop a dozen report requests into the queue and have finished chats ready by the start of the day.
- Generator: paste a list of lines (product names, for example) and a sample — a task is created for each line, following the sample;
- Maximum active chats limits how many tasks the agent runs at once, so the model isn’t overloaded;
- Auto-switch: when a new task starts, the chat panel jumps to it on its own;
- Peak hours: the setting controls whether model prompts are sent during the models’ peak load hours or held back until the dip;
- the queue is shared — it is the same in Analytics Agent and in Xedant Agent’s own interface; changes on either side are visible everywhere at once.
What Xedant Agent is and how to install it — on the Xedant Agent page.