First Agent is the first agent I call on when I need help with a code problem.
Custom built based on the results of building the agent in Boot.dev, this agent is being customised to work exactly the way I want it to, with the focus being on assisting me learn rather than getting code out as fast as possible.
- Finding files within a directory
- Reading files within a directory
- Running Python code
- Not being able to write (write_file has been commented out of call function, and exists only so I can reimplement if I need to quickly)
- Running on my local hardware using Ollama endpoint.
As I run different configurations for this, is local.py. That has a few variables:
- MODEL, which is where the selected backend model is selected; Gemma4:12b on my main system, Qwen3.5:9b on my laptop. These seem to give a good balance of being able to run tools and provide the intended style of output, while running at a fast or moderate speed for the hardware I'm using. (Note: Qwen3.5:9b, and the one I used before that Qwen3.5:4b, has started bugging out, at least on my laptop, and not using ending up giving a response. This seems to be a limitation of token count on my laptop, as it seems to start having issues around 4k tokens total
- SERVER, which is used for my most powerful instance of Ollama. The code checks whether this is available. On my desktop, for now, this is kept the same as the next variable. Point this wherever your most powerful Olama (or AI endpoint of your choice) is.
- LOCAL, which is used for the local Ollama endpoint on the device. On my laptop, this is primarily a fallback measure for when I'm not at home, or my desktop is out of commission. This is intended to be an endpoint that shouldn't fail.