llmterm

A minimal terminal client for OpenAI-compatible LLM servers.

llmterm is a small Python command-line application for interacting with local or self-hosted LLM servers that expose an OpenAI-compatible API.

Features

• Interactive endpoint selection
• Automatic model discovery
• Interactive model selection
• Asynchronous streaming responses
• Conversation history
• Optional custom system prompt
• Optional Markdown conversation logging
• .env support for API keys
• Small, modular implementation

Project structure

llmterm/
├── pyproject.toml
├── README.md
├── .gitignore
└── src/
    └── llmterm/
        ├── __init__.py
        ├── endpoint.py
        ├── main.py
        └── utils.py

src/ contains only package source code. Runtime-generated files are not stored inside the Python package.

Markdown output location

When Markdown saving is enabled, responses are stored in:

~/.llmterm/
└── markdown-outputs/
    ├── model-name-08-15-2026.md
    └── another-model-08-15-2026.md

This location is independent of the directory from which llmterm is run.

Keeping generated data outside the package is important for a distributable Python package because installed package directories should not be assumed to be writable.

Installation

Using uv

Install llmterm-prasannaba as a standalone CLI tool:

uv tool install llmterm-prasannaba

Run:

llmterm

Using pip

You can also install llmterm-prasannaba using pip:

python -m pip install llmterm-prasannaba

Run:

llmterm

Development

Clone the repository and enter the project directory:

git clone https://github.com/prasannaba/llmterm.git
cd llmterm

Install the development environment and dependencies:

uv sync

Run the application:

uv run llmterm

Optional API key

Create a .env file in the project directory when Unsloth Studio requires an API key:

UNSLOTH_STUDIO_API_KEY=your-key

Configured endpoints

The corresponding server must be running and expose an OpenAI-compatible API.

Architecture

Terminal
   │
   ▼
llmterm
   │
   ▼
OpenAI Python client
   │
   ▼
OpenAI-compatible /v1 endpoint
   │
   ▼
Local/self-hosted LLM

endpoint.py

Owns the asynchronous OpenAI-compatible client, model discovery, chat completion calls, and client cleanup.

main.py

Handles endpoint selection, model selection, conversation history, streaming, and API errors.

utils.py

Handles configuration, system prompts, response cleanup, and Markdown persistence.

The project intentionally avoids a large agent framework or abstraction layer.