Xedant Analytics Agent

Fast data analysis and visualization: dashboards you can dig into deeper than Grafana

Xedant Analytics Agent was built by an analyst with more than 25 years of working with data and visualization — it is a tool for real analysis, not just monitoring. Here you don’t only look at dashboards: you dig into the data — filter, cross-check, calculate in Python. You can ask the AI to add reports, charts and tables in plain language, right inside the interface. Ready-made solutions — forecasts, segments, scoring — come from a library of 257 analysis templates, including a free analytics course. And when something goes wrong, the system tells you itself: alert rules check your data on a schedule and send a signal to Telegram, Slack or email. The data for reports is usually prepared by Data Agent — together the products form a pipeline: collect → prepare → present.

Documents as reports built from blocks: charts, tables, metrics

38 block types — from Markdown to Jupyter Notebook

28 chart types: lines, bars, funnels, region maps, relationship diagrams

SQLite, PostgreSQL, ClickHouse and your own Python scripts out of the box

Schema auto-detection and caching of tables and queries

Lakehouse: queries to Parquet files without configuring sources

Interactive tables with pagination and cell links

Template library: 257 ready-made analysis solutions with sample data

Ask the agent to add a report, chart or table

Fix with AI button next to any error

Runs on your central Xedant Agent — no extra keys or subscriptions

Built-in validation of documents, blocks and SQL queries

Auto-fix: errors are visible before a report is published

Jupyter Notebooks: Python inside a report

pandas, NumPy, Matplotlib, Seaborn for data analysis

External Python scripts extend the pipeline: source, transform, alert, delivery

Scriban templates and variables in reports

One sidebar and agent chat on every page

Real-time report updates via SignalR

Reports are JSON files: Git, versioning, instant reload

13 types of block styling overrides

Public reports and access control

English and Russian interface, chosen in settings

Self-hosted: one Docker container, data in persistent volumes

Created by an analyst with more than 25 years of experience

Alerts: rule checks and notifications in 24 channels — from Telegram to email

Does not load the server when idle: background checks wake on changes only, and the search index lives on disk without holding memory

Pictures and files inside a report — a picture is shown in place, a file downloads from its link

Reports built from blocks

Store Pulse report built from blocks: four KPI tiles, monthly and weekly revenue charts, a channel table and a verdict tile on one canvas

A document is a grid of blocks from 1 to 24 columns wide

Drag and drop blocks with the mouse

Reports are stored as JSON files — version them in Git

Changes are picked up instantly: real-time via SignalR

Filters, what-if parameters and click-to-drill right in the report

Export a report as PDF — send it to a colleague or a client

Thirty-eight block types

Ops Console report mixing block types: metric tiles, a bar-and-line pulse chart, a weekday comparison line chart and a data table in one document

Charts, tables, metrics and gauges

Markdown text and section headings

Bookmark links and entity property lists

Jupyter Notebooks as full-fledged report blocks

Charts for any data

Funnel Leaks report: conversion KPI tiles, a monthly step-rate chart with three series and a month-by-month funnel table

Line, bar, scatter and doughnut charts

Distribution histograms and box plots with quartiles

Financial OHLC candlesticks with trading volume

Series, stacks, smoothing, legends and animations

Funnels, Gantt, flow, relationship diagrams and region maps

Connecting to your databases

The Data Sources list: SQLite connections with type badges, a filter box and an Add Data Source button

SQLite, PostgreSQL, ClickHouse and your own Python scripts

Automatic schema detection for data sources

Caching of tables and query results

Queries are validated before execution — errors surface immediately

AI right inside the report

The agent chat opened right on the Store Pulse report: the message is pre-bound to the report's document.json, with model and skill pickers

“Add a weekly sales chart” — and the block appears

New reports, charts and tables in plain language

Fix with AI button next to an error

The agent edits the report directly — updates are instant

Analysis template library

The 257-template library: the Analytics 101 course, Business and Finance groups, each template with what it gives, tags and a demo dataset

257 ready-made templates in two groups — Business and Finance — plus a course

Forecasts, segments, scores, anomalies and alerts — end to end

Demo data with every template — run and study before adapting

“Use this template” — the agent adapts it to your data

A free 50-lesson course on data analysis is part of the library

Validation and auto-fix

An agent chat that validated every analytics document — static, then live — and fixed what it found: the task list and the final fixed-and-verified report

Document validation: blocks, IDs, widths, configurations

SQL query validation before execution

Background validation of all reports with result caching

Errors are visible before publishing — with the AI fix button

Jupyter Notebooks

A template report page with its How to run section: the Python scripts (extract.py, analyze.py, load.py) that build the report's data

Python inside a report: execution via papermill

pandas, NumPy, Matplotlib, Seaborn out of the box

Automatic environment and dependency installation

Templates: basic, data analysis, machine learning

Extending with your own scripts

The Scripts page: external Python scripts hook into the pipeline — none configured in this demo yet; script definitions are JSON files that appear here immediately

Your own data source out of any service with a Python client

The script re-works the query result: rename, convert, enrich, compute

A custom alert condition and a custom notification channel

Scripts are plain files next to your settings — no rebuild, no restart

Lakehouse

The lakehouse file browser opened on orders.parquet: 26,500 total rows with the schema and first 20 rows previewed right in the browser

A file browser for all your data

Queries to Parquet files via Apache Arrow

SQLite files and Markdown without configuring sources

Table schemas and samples right from the browser

Variables and templates

Document Settings dialog: range and timeframe selectors, the drill filter, filter variables and what-if parameters that report queries bind as variables

Dynamic reports on the Scriban engine

Variables in document paths and content

Date range and timeframe selectors

One template — as many reports as you need

Styling without limits

Supplier Scorecard report: an on-time-delivery bullet chart with threshold bands and a table with formatted currency, percent, days and payment terms

13 types of block styling overrides

Number colors by thresholds, formats with prefixes and suffixes

Renaming columns and links from table cells

Hiding series, line, bar and point styles

Alerts — the system watches for you

Alert rules in the RoamReady store watch group, evaluated every 1d: firing and normal rules with severity labels, evaluation times and instance counts

A rule is a query with a condition: sales fell below a threshold — you get a signal

Scheduled checks: from every 10 seconds to once a day

24 notification channels: Telegram, Slack, email, webhook and more

Silences and mute timings: no noise at night, on weekends and holidays

Firing history and a notification log — who got what and when

Recording rules: check results are saved into a database — build your own dashboards on them

The assistant agent and the sidebar

The Chats dialog over the app: conversations with message counts, costs, commits and the model and skill used for each — one sidebar on every page

One sidebar on every page: sections, documents and search

Agent chats open from anywhere — no switching between apps

Give a task in plain language: add a chart, fix an error, prepare a report

A prompts queue with auto-send: the agent works through tasks one by one

Pick a model and a skill for every request — the analytics skill is pre-selected

As-you-type search across documents and templates, and your recent history