Data Agent is an agent for collecting and preparing data. You describe in chat which data you need — the agent writes and maintains Python scripts on its own, runs them on demand or on a schedule, and puts the results into the Lakehouse store. Data Agent serves as the data source for Analytics Agent: everything it collects and prepares, Analytics Agent shows in reports and charts. The ETL — extracting, transforming and loading data — is done entirely by agents: there is no manual data entry and no script editing in the interface.
Who it is for
For business owners and teams who need regular data: competitor prices, order statistics, exports from external service APIs. Before, this meant hiring a developer or figuring out exports yourself. Here it is enough to explain to the agent what you need — it writes the code, sets the schedule and keeps the export working.
System requirements
You only need a Linux server with Docker. The Python environment with all the data libraries is built into the image — nothing to install separately. The detailed walkthrough is on the Data Agent Installation page.
First sign-in — self-registration
The login and password are not set in advance: on first launch the application is open for registration, and the first account created becomes the administrator. After that, registration closes. Sign-in works through a session token (90 days), passwords are stored as hashes — they cannot be read from the database.
Interface sections
- Home — an overview: runs and harvests today, the instance’s state, quick links into every section;
- Chats — talking to the agent: this is where you ask for data and fixes in plain language. The section appears when
AGENT_API_URLandAGENT_API_KEYare set (the address and key of your agent); - Scripts — the tree of folders and files the agent wrote: every folder has a description, every file has a page with its code;
- Runs — the history of all runs with live output and logs;
- Scheduler — the run schedule: the agent manages it at your request; here you see what runs and when, with a “Run now” button;
- Lakehouse — the store of collected data, laid out by project and source: Parquet, databases, descriptions, logs;
- Data Sources — the registry of sources you can query: they appear automatically;
- Harvesters — data collectors: services for bulk internet search, page download and text processing by models;
- Harvests — bulk collection jobs: what the agent launched, how much is processed and what it cost;
- Analytics — an overview of the agent’s work: tokens, cost, load by hours and days;
- Settings — theme, language, font size, the Python interpreter for runs;
- Mobile — a mobile web app for iOS and Android: the same capabilities on your phone.
The dark theme is on by default; the interface language is English (each user picks their own in settings).
Your first script in five minutes
- Describe to the agent in chat which data you need and where it comes from (for example: “Every day, collect competitor prices from the catalog”).
- The agent creates a source folder in the scripts folder: the export script
extract.py, the database build scriptload.pyand a description file. - Ask it to run — the data appears in the Lakehouse, and the source becomes available automatically.
- Ask it to set a schedule — from then on the data refreshes on its own.
- Connect the source in Analytics Agent — and build reports and charts on fresh data.
Further in the documentation
- Chat with the Agent — how to ask for tasks, models and the message queue;
- Scripts — how the scripts folder is arranged and what lives in it;
- Runs — running scripts manually and on a schedule;
- Scheduler — the run schedule and its rules;
- Lakehouse — the data store: what lies where;
- Data Sources — how data becomes ready for queries;
- Harvesters — bulk data collection: search, page download and text processing by models;
- Data for Analytics Agent — the collect → prepare → present pipeline;
- Agent Analytics — tokens, cost, load;
- Xedant Agent Integration — how the pairing with the central agent works.