How to Get Started with MindsDB Anton

MindsDB Anton is an open-source autonomous BI agent that connects to your data sources and answers analytical questions by planning, writing, and executing its own SQL and Python code. Unlike chat-based analytics assistants that stop at generating a query you still have to run yourself, Anton takes ownership of the entire workflow. It plans the analysis, writes the code, executes it inside an isolated environment, and returns a finished output including charts, tables, and explanations. Released in April 2026, Anton was downloaded over 10,000 times in its first week, which suggests real demand for this class of tool among analysts and ops teams who want answers, not more SQL to review.

What Anton Actually Does

Most analytics tools require you to do at least half the work. You ask a question, the tool writes a query, and you run it, clean the result, and build a visualization. Anton compresses that entire loop into one prompt. Ask "Which product categories drove the most revenue growth last quarter?" and Anton will write the SQL to pull the data, execute it in its own sandbox, detect any errors and self-correct, then return a finished answer with a chart.

The core mechanism is the scratchpad, an isolated execution environment Anton spins up for each task. Think of it as a private notebook that only Anton sees during the analysis. You can ask Anton to dump the scratchpad at any point and receive a full notebook-style breakdown of every code cell it ran, every output, and every error it caught and fixed. This auditability separates Anton from black-box chatbot tools where you receive an answer but have no way to verify how it was produced. In business settings where finance or ops teams will challenge a number, showing the reasoning trace matters.

Anton is MIT-licensed and runs entirely on your local machine. Your data does not leave your computer. Credentials are stored as environment variables on your system, not passed to the underlying language model. This makes it viable for companies with data residency requirements.

What You Need Before You Start

You need macOS or Linux (Windows via WSL is supported but not yet fully stable as of April 2026). You need an account with one of the supported LLM providers: Minds Cloud, OpenAI, or Anthropic. You also need at least one data source to connect, which can be a local CSV file, a PostgreSQL or MySQL database, a REST API, or any of the 200-plus sources Anton supports out of the box.

No Python or Node.js installation is required. Anton ships as a self-contained binary. The total install footprint is under 50MB.

Step 1: Install Anton

Open a terminal and run the following commands:

curl -sSf https://raw.githubusercontent.com/mindsdb/anton/main/install.sh | sh
export PATH="$HOME/.local/bin:$PATH"

This installs the Anton CLI. If you prefer a visual interface, the Anton Desktop App for macOS is available from the MindsDB website and shares the same configuration and data source connections as the CLI. Both interfaces run the same underlying agent.

Verify the install with:

anton --version

Step 2: Configure Your LLM Provider

On first launch, Anton prompts you to choose a language model provider. This is the model that will plan analyses and generate code. Three options are available.

Minds Cloud is the recommended choice. It is purpose-built for Anton and automatically routes each task to the best available model depending on complexity. Create an account at mdb.ai, select a plan, generate an API key, and paste it when prompted. Minds Cloud handles model selection automatically, which removes one variable from troubleshooting if something goes wrong.

OpenAI and Anthropic are both supported if you already have API keys. GPT-4o and Claude 3.5 Sonnet both perform well for data analysis tasks. Paste your key at the provider prompt during setup.

You can switch providers at any time by running anton config.

Step 3: Connect a Data Source

Anton supports two connection modes. The secure connection mode, recommended for anything with authentication, collects your credentials and stores them as environment variables on your machine. The underlying language model never receives the raw credentials. The direct connection mode is for public APIs that require no authentication.

To connect a PostgreSQL database, type:

/connect postgres

Anton will prompt for host, port, database name, username, and password, then run a test query to confirm the connection.

To connect a local CSV file, type:

/connect csv

Point Anton at the file path. It will read the schema, display column names and sample rows, and confirm it parsed the file correctly before you start querying.

For sources not on the built-in list, select "Custom datasource" and describe what you want to connect. Anton will generate a connection template from your description, prompt you for the required parameters, and validate the connection before saving it.

Once a source is connected, you can manage it using commands to list all connected sources, update credentials, or disconnect a source entirely.

Step 4: Ask Your First Question

With a data source connected, ask a question in plain English. No SQL knowledge is required.

What were the top five revenue-generating customers last month?

Anton responds by writing SQL or Python to retrieve and aggregate the data, executing the code in its scratchpad, handling any errors automatically, and returning a formatted answer. Most common failures (wrong column names, type mismatches, null values in aggregations) are caught and corrected without any input from you.

To see the full reasoning trace, run:

/scratchpad dump

This prints every code cell Anton ran for the last task, with inputs, outputs, and error messages, in the order Anton executed them. You can copy specific cells and run them independently if you want to modify the analysis.

Step 5: Export and Reuse Outputs

Anton can export results as CSV, PNG, or a shareable HTML report. After an analysis, run:

/export last

Anton asks what format you want and saves the output to a local directory. For questions you run repeatedly, you can save them as named reports that re-execute against fresh data each time. This is the simplest way to build a lightweight recurring report without setting up a full data pipeline.

Common Workflows

Sales and revenue analysis is the most common starting point. Connect a CRM export or a production database and ask which reps closed the most deals in a period, what the average deal size is by industry, or which pipeline stage has the highest drop-off rate.

Operations teams use Anton to query production databases for failure rates, processing times, and anomalies. A typical prompt: "How many orders failed processing in the last 24 hours, and what was the most common error reason?" Anton returns a breakdown without requiring anyone to write the underlying query.

Finance teams connect expense spreadsheets or accounting exports and ask for month-over-month breakdowns, budget variance summaries, or trend reports by department.

In each case the output is a finished artifact ready to share, not a starting point for more manual work.

Current Limitations

Anton does not yet support real-time streaming data. Scheduled report automation requires an external scheduler such as cron; there is no built-in scheduling in the April 2026 release. Windows support without WSL is not yet available. Very large datasets (above several hundred million rows) may require partitioning the query manually before passing it to Anton.

Summary

Anton installs in under two minutes, connects to any major database or API without configuration files, and handles the full path from question to finished output. The scratchpad architecture gives every answer an auditable reasoning trace, which makes it practical for teams where numbers get challenged. For recurring data pulls, one-off investigations into live databases, or operational monitoring without a dedicated analyst, it removes the manual steps between question and result.