Agent

About

Agent is a simple wrapper for enabling agentic systems using the OpenAI and Llama.Cpp REST APIs.

Note: This is an experimental, personal toy project. Not intended for production use—expect rough edges!

Setup

You’ll need to clone the project for now:

git clone https://github.com/teleprint-me/agent.git agent
cd agent

You can also use pip to install the package locally:

pip install git+https://github.com/teleprint-me/agent

Python 3.12.x is recommended for stability and compatibility with most ML tooling. This lets other tools catch up to Python 3.13+ while keeping your environment reliable.

python3.12 -m venv .venv
source .venv/bin/activate
pip install --upgrade pip
pip install -r requirements.txt

• Tip:

  • Avoid pinning dependencies during early development.
  • For a stable setup, freeze requirements with: pip freeze > requirements.lock.txt

Dependencies

Dependencies are intentionally minimal:

• Python 3.12+ — Stable baseline
• OpenAI and gguf — Model compatibility (OpenAI + local LLMs)
• Prompt Toolkit and Pygments — CLI and theming for the console
• Tkinter and ttkbootstrap — GUI and theming for the editor
• jsonpycraft, dotenv, peewee, etc. — Planned config and utility tools

Other utilities are included but not fully integrated yet.

Usage

If using the llama.cpp server backend, ensure you start it with the correct flags to enable templates, functions, and pooling:

llama-server \
    --port 8080 \
    --n-gpu-layers 32 \
    --ctx-size 16384 \
    --pooling mean \
    --slots --jinja -fa \
    --reasoning-format none \
    --reasoning-budget -1 \
    -m /path/to/ggml-model-f16.gguf

Set up a .env file for your API connection:

touch .env

Add the following environment variables:

OPENAI_API_KEY=sk-no-key-required
OPENAI_BASE_URL=http://localhost:8080/v1

Launch

• CLI:

  python -m agent.cli

The text user interface should appear.

• GUI:

  python -m agent.gui

The editor window should appear.

• Config:

A simple CLI tool is included for customizing agent and model settings. For help:

python -m agent.config -h

Configuration is managed via jsonpycraft for clean, editable JSON-based settings.

Implementation Notes

• The GUI is currently minimal—just a styled window with basic file actions (open, close, new, save, save as) and simple tabs. No real “agent” features yet; it’s a skeleton to build on.
• The CLI is getting most of the early development attention. Once the CLI matures, focus will shift to the GUI.
• The priority is to make the core API as agnostic as possible, so both the CLI and GUI can share the same backend logic.
• Current work is focused on the basics: streaming, tool calling, reasoning, configuration, chat loop cycle, and simple input.

It’s a moving target. Expect frequent changes, bugs, and some questionable design decisions (at least until it all clicks together).

Resources

• llama.cpp
• OpenAI API
• jsonpycraft
• prompt-toolkit
• Tkinter
• ttkbootstrap

License

Licensed under the AGPL to ensure the code remains free and publicly available. This guarantees user freedom while protecting against closed-source forks.