How to Set Up MindsDB Anton for Data Analysis

MindsDB Anton launched on April 2, 2026, as an open-source autonomous business intelligence agent. You describe what you want to understand about your data, and Anton pulls the data, writes and executes code, builds charts, and returns a complete analytical output. This guide walks through installation, LLM provider configuration, and your first working query. No SQL or Python knowledge is required.

What Anton Does Differently

Most BI tools are query interfaces. You write SQL, configure a report, or drag fields into a chart builder. Anton inverts that model: you describe an outcome in plain language, and Anton plans and executes the steps to reach it.

When you send Anton a question, it breaks the problem into steps, selects which data sources and tools to use, writes code, runs it in an isolated execution environment, and returns a result. The output can be a table, a chart, an interactive HTML dashboard, or a written explanation of what the data shows. Anton retains memory across sessions, so it learns your data structure and improves over repeated use.

MindsDB, the company behind Anton, has raised over $50 million from Benchmark, Mayfield, Y Combinator, and NVIDIA, and its tools are used by hundreds of thousands of developers. Anton itself reached 535 GitHub stars in its first week, with the codebase comprising 751 commits across 9 releases as of April 2026.

What You Need Before You Start

Before installing Anton, confirm you have:

• macOS or Linux (Windows support is in beta via a PowerShell installer; a native Windows app is listed as coming soon)
• Python 3.11 or higher (Anton bootstraps its own virtual environment, but Python must already be present on the machine)
• An LLM API key from one of three sources: Minds Cloud ($35 per month starting plan), OpenAI, or Anthropic. Anton uses Anthropic Claude as its default model. If you already have a Claude or GPT-4 API key, you can start immediately.
• curl (standard on macOS and Linux)

No database admin access is required to get started. Anton works with a local CSV file the same way it works with a production Postgres database.

Step 1: Install Anton

Open a terminal and run:

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

This installs the Anton CLI and adds it to your PATH. When the install completes, run:

anton

Anton launches and prompts you to configure your LLM provider. First-run setup takes about 90 seconds.

Mac users who prefer a desktop application can download the signed installer directly from the GitHub releases page at github.com/mindsdb/anton/releases. The desktop app provides the same functionality as the CLI with a graphical interface.

Step 2: Choose Your LLM Provider

Anton presents three options for the AI engine that powers its analysis:

Minds Cloud is the recommended option if you want everything pre-configured. Create an account at mdb.ai, choose a plan, generate an API key, and paste it into Anton when prompted. Minds Cloud is purpose-built for Anton with no additional setup. The starting plan is $35 per month for 5 million tokens. MindsDB is offering one month free to the first 1,000 signups using the code MDBCOMMUNITY100.

OpenAI or Anthropic work directly if you already have API keys. Anton accepts them without additional configuration and stores them in its local credential vault. Keys are never passed as part of LLM prompts. Only the key name is visible to the model during reasoning. The actual value stays encrypted in the vault.

After you enter a key, Anton confirms the connection and proceeds to the main interface.

Step 3: Connect a Data Source

From the main Anton prompt, type /connect to open the credential vault and add a data source. Anton walks you through connection templates for each supported source.

For a PostgreSQL database, it asks for host, port, database name, username, and password. For a CSV file, you provide the local file path. Anton reads the schema on first connection and stores column names, data types, and table relationships in its memory layer so it can reference them in future conversations without you needing to re-explain the structure.

Supported data sources include PostgreSQL, MySQL, Microsoft SQL Server, Oracle Database, MariaDB, Amazon Redshift, Google BigQuery, Snowflake, Databricks, Salesforce, HubSpot, Shopify, and any custom source accessible via REST API or MCP.

If you have multiple data sources, connect all of them during setup. Anton can join across sources in a single analysis: pulling Salesforce pipeline data alongside a Postgres orders table, for example.

Step 4: Ask Your First Question

With a data source connected, return to the main prompt and type a question in plain language:

What were the top 10 products by revenue last month, and what changed compared to the month before?

Anton responds by planning the analysis (it shows you the steps before executing), writing the SQL or Python needed, running the code in an isolated sandbox, and returning a formatted table with a written explanation of what changed and why.

For more complex requests, Anton builds a self-contained HTML dashboard you can open in any browser and share with stakeholders. The dashboard does not require Anton to be running to view.

Follow-up questions preserve context from the session:

Break that down by region and flag anything that looks anomalous.

Anton remembers which products and time range were referenced and extends the previous analysis without restarting from scratch. This is where the memory system becomes visible: it retains session context, stores analytical lessons learned across sessions, and archives past conversations with timestamps for retrieval.

What Anton Returns

Depending on what you ask, Anton produces one or more of:

• Tables: Structured query results formatted for readability
• Charts: Bar charts, line graphs, and scatter plots rendered inline
• HTML dashboards: Exportable, interactive multi-panel reports
• Written explanations: Plain-language summaries of what the data shows and what is likely driving it
• Next-step suggestions: Anton identifies patterns and recommends follow-on questions

The analytical scratchpad for each session is saved and reproducible. A colleague can re-run or extend the analysis from the same scratchpad without needing to repeat your setup.

Limitations to Know

Anton works best with structured, tabular data. It does not process unstructured text documents or image files. For CSV files larger than several hundred megabytes, performance depends on the LLM context window.

First-session queries against an unfamiliar schema may occasionally produce errors as Anton infers column relationships. Correcting it once is usually sufficient for the remainder of the session, and it stores that correction in semantic memory for future sessions.

Windows support is in beta. The PowerShell installer exists and works, but the native Windows desktop app is listed as coming soon on the GitHub releases page.

For teams that prefer a browser-based option with no installation, VSLZ AI handles data analysis from a file upload with no local setup required.

Practical Starting Points

Three questions that surface immediate value for common business roles:

1. Sales operations: "Summarize pipeline by stage and flag deals that have been stuck for more than 30 days."
2. Product teams: "Show weekly active users over the past 90 days and highlight any weeks with unusual drop-offs."
3. Finance: "Break down operating expenses by category for Q1 and compare to Q4 last year."

Each produces a multi-part output that Anton refines in follow-up turns without re-specifying the data source or time range.

Anton represents a different model for business intelligence tooling. Rather than building reports ahead of time and hoping they answer the right question, you describe the question and receive a complete answer including the analysis, charts, and explanation. For teams that handle frequent ad-hoc data requests, that shift reduces the overhead substantially.