Skip to main content

Quickstart

Anton is an open-source AI coworker that can execute tasks, connect to tools and data, remember lessons, and improve its workflows over time. This page takes you from nothing to a working Anton — and shows you the learning loop that makes it different from any other agent — in about five minutes.

1. Install Anton​

macOS / Linux:

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

Windows (PowerShell):

irm https://raw.githubusercontent.com/mindsdb/anton/main/install.ps1 | iex

Prefer a desktop app? See Desktop app. For offline installs, Linux distro notes, and troubleshooting, see Installation.

2. Run it​

anton

On first run Anton shows its terms, then asks you to pick an LLM provider.

3. Pick a provider​

The recommended default is Minds (mdb.ai) — choose option 1. Minds gives you smart model routing, cost optimization, and secure data connectors with a single API key. If you don't have a key yet, Anton opens the signup page for you; it takes a few seconds.

Already have an Anthropic, OpenAI, or Gemini key?

Choose option 3 (Bring your own key), then pick your provider and paste your API key. Anton validates the key with a quick probe call before saving it. You can also point Anton at any OpenAI-compatible endpoint — Ollama, vLLM, Together, Groq, and friends. The full comparison lives in Pick a provider, and you can switch any time with the /llm command.

Your key is stored in ~/.anton/.env, so it carries across sessions and workspaces.

4. Ask Anton to do something​

Try a small local task first — no connections or extra keys needed:

you> Write a Python script that prints the first 10 Fibonacci numbers, and run it.

Watch what happens: Anton opens a scratchpad — an isolated code execution environment — writes the script, runs it, and shows you the output. That scratchpad is the core of how Anton works: most tasks, from web scraping to database analysis to building dashboards, run through it. You describe the outcome; Anton writes and executes whatever code gets there.

5. Save what it learned as a skill​

This is where Anton stops being a stateless chat agent. Tell it to remember the procedure it just performed:

/skill save fibonacci

Anton reads the recent work, drafts a step-by-step procedure, and saves it to its skill library at ~/.anton/skills/. Next time a request matches, Anton recalls the procedure instead of reasoning from scratch. More in Skills.

6. Peek at Anton's memory​

Anton has been taking notes the whole time. Look inside the workspace folder it created:

ls .anton/

You'll find, among other things:

  • memory/lessons.md — facts Anton learned while working
  • memory/rules.md — behavioral rules (always / never / when)
  • episodes/ — a timestamped log of this session

These are plain markdown files — open them, read them, edit them. You can also inspect everything from inside the chat with /memory. See What Anton remembers.

7. Resume the session​

Quit (exit or Ctrl+D), then start Anton again and pick up where you left off:

/resume

Anton lists your previous sessions and restores the conversation. Between the skill you saved, the lessons on disk, and the resumable session, you've now seen the loop that defines Anton: work → learn → improve.

Next steps​