DeepChat

DeepChat is a small Python CLI that launches an Ollama-backed chat session. It supports two runtime modes:

• Local mode: uses the Ollama server installed on your machine.
• Container mode: starts the official ollama/ollama Docker image and talks to it over http://localhost:11434.

On startup, the script prepares the selected runtime, pulls the requested model, runs a quick test prompt, and then drops into an interactive prompt loop.

Requirements

• Python 3
• Internet access for the first model pull
• For local mode: Ollama installed and available on your PATH
• For container mode: Docker installed and running

Usage

python main.py [model] [--local|-l]

• model is optional. If omitted, the script uses deepseek-r1:1.5b.
• --local or -l is optional.
• If no flag is provided, the script uses Docker mode.

Modes

Local mode

python main.py deepseek-r1:7b --local
python main.py deepseek-r1:7b -l

Local mode:

• upgrades pip
• installs the Python ollama package if needed
• starts ollama serve
• pulls the requested model locally
• opens an interactive chat session through the Python Ollama client

Docker mode

python main.py deepseek-coder-v2

Docker mode:

• installs the Python requests package if needed
• stops and removes any existing container named ollama
• starts a fresh ollama/ollama container
• pulls the requested model inside the container
• opens an interactive chat session through the Ollama HTTP API

Examples

# Use the default model in Docker mode
python main.py

# Run a local Ollama model
python main.py deepseek-r1:7b --local

# Run a model in Docker mode
python main.py deepseek-coder-v2

Large Context Inputs

If you want to paste larger prompts or code blocks, you may need a model variant with a higher context window. This repository includes a modelfile that sets:

• num_ctx 24576
• num_predict 8192

Create a patched model like this:

ollama pull deepseek-coder-v2
ollama create deepseek-coder-v2-patch -f modelfile
python main.py deepseek-coder-v2-patch --local

You can replace deepseek-coder-v2 with any compatible base model.

Notes

• The script rejects unknown flags and exits if too many arguments are provided.
• In Docker mode, the container is always named ollama.
• The interactive output strips <think>...</think> content before printing visible answers.

Screenshot