The goal of this meetup is to help you get started using AI agents for free.
By the end of this meetup you'll have a workflow that will take you a long way before you feel the need to spend any money.
Anyone who writes code by hand and wants to try an agentic AI workflow for free, e.g., data scientists and software developers.
A lot of people are now using AI agents to write some or all of their code. One of the best AI agents today is Claude Code, but using it for real work can cost a lot. This meetup shows a good and free alternative so you can practice without worrying about cost.
- Understand what an AI agent is.
- Install and get started with a good and free AI agent.
- Set it up to use R and Python interactively.
- Use it inside your IDE.
An AI agent is a large language model (LLM) plus additional software (harness) that allows it to act on your behalf.
By itself, the LLM can guide you, but the one who takes action is still you.
ollama run qwen2.5:7b>>> Write a new empty file.txt
To create an empty file named `file.txt` on a Unix-like system (such as Linux or macOS), you can use the following command in the terminal:
```bash
touch file.txt
```
With a harness, the same LLM can now act on your behalf.
opencode## User
Write a new empty file.txt
## Assistant
**Tool: write**
## Assistant
Created an empty file named file.txt in the current directory.
There are many agents and models. Here we'll focus on a few that are useful and free.
What is OpenCode?
OpenCode is an open-source agent that helps you write and run code with any AI model.How much does OpenCode cost?
OpenCode is 100% free to use. It also comes with a set of free models. There might be additional costs if you connect any other provider.
You can try opencode in a Docker container with:
docker run -it --rm ghcr.io/anomalyco/opencode
Or install it following the instructions for your OS, for example:
curl -fsSL https://opencode.ai/install | bash
Launch it:
opencode
Type / to see everything you can do. For example, use /models and /connect to switch to a provider like Google with an API key from Google AI Studio.
Coding agents typically run R or Python code with Rscript -e or python -c.
[This] keeps forcing the agent to rebuild context. mcp-repl keeps the session open instead: variables, loaded packages, plots, and other state stay available until you or the model reset.
-- https://github.com/posit-dev/mcp-repl
Now you can ask an AI agent to "read https://github.com/posit-dev/mcp-repl and set up mcp-repl for opencode."
These examples show the behavior with and without that tool:
- With mcp-repl enabled
## User
Write a new empty file.txt
## Assistant
**Tool: write**
## Assistant
Created an empty file named file.txt in the current directory.
- With mcp-repl disabled
## User
Run x <- 1 in R
## Assistant
Running `x <- 1` in R.
**Tool: bash**
## Assistant
Executed `x <- 1` in R via `Rscript`:
```
[1] 1
```
`x` assigned value `1` successfully.
## User
Now run exists("x")
## Assistant
Running `exists("x")` — noting Rscript sessions are isolated, so chaining with previous assignment.
**Tool: bash**
## Assistant
`Rscript -e 'print(exists("x"))'` returns:
```
[1] FALSE
```
Each `Rscript -e` invocation is an isolated process, so `x <- 1` from the previous call does not persist.
In the same session:
```
Rscript -e 'x <- 1; print(exists("x"))'
[1] TRUE
```
OpenCode, like most AI agents, can run in any terminal, including inside IDEs like RStudio or Positron.
Now go ahead, build something useful and have fun!