OpenAI Simple UI

A minimal Streamlit-based chat UI for OpenAI models (example list: gpt-5, gpt-5-mini, gpt-5-nano). It lets you adjust reasoning_effort and verbosity from the sidebar and streams tokens as they arrive.

Features

• Very small chat interface using st.chat_message
• OpenAI Python SDK v1 streaming (client.chat.completions.create(stream=True, ...))
• Sidebar controls for reasoning_effort (high / medium / low / minimal) and verbosity (high / medium / low)
• Easy to extend the model list
• Runs locally or via container (Podman / Docker)

Project Structure

.
├── Dockerfile
├── build.sh              # Build image (tags with current date)
├── container.sh          # Example Podman run script
├── script.sh             # Local run script
├── template_container.sh # Template for container run
├── template_script.sh    # Template for local run
└── src/
    ├── openai-simple-ui.py
    ├── requirements.txt
    └── run.sh            # Startup script (custom baseUrlPath etc.)

Screenshot

Prerequisites

• Python 3.12+
• An OpenAI API key (OPENAI_API_KEY environment variable)

Local Run

git clone <repo>
cd openai-simple-ui
python -m venv .venv
source .venv/bin/activate
pip install -r src/requirements.txt
export OPENAI_API_KEY=sk-...  # Your key (do NOT commit it)
streamlit run src/openai-simple-ui.py --browser.gatherUsageStats=false

Container Run (Podman example)

./build.sh                 # produces image: openai-simple-ui:YYYYMMDD
export OPENAI_API_KEY=sk-...
podman run -d \
  -p 8511:8501 \
  -e OPENAI_API_KEY=${OPENAI_API_KEY} \
  --name openai-simple-ui \
  openai-simple-ui:YYYYMMDD

Use docker instead of podman if you prefer Docker.

src/run.sh sets:

--server.baseUrlPath /simple-openai
--server.address 10.0.2.100
--server.port 8501

Remove or edit those flags if you don’t need a custom base path or address binding.

Environment Variables

Name	Description	Required
OPENAI_API_KEY	OpenAI API key	Yes

Example .env file:

OPENAI_API_KEY=sk-...

Load it (bash):

export $(grep -v '^#' .env | xargs)

Code Outline

1. Stores conversation in st.session_state["messages"] (role + content)
2. On user input, appends the user message
3. Streams assistant response token-by-token and updates the placeholder

Model list lives here:

model_options = ["gpt-5", "gpt-5-mini", "gpt-5-nano"]

Add or remove entries as needed.

Dependency Management

Pinned versions in src/requirements.txt. To add a package:

pip install <package>
pip freeze > src/requirements.txt   # optionally prune unrelated packages first

Customization Examples

Requirement	How
Add model	Append to model_options list
Remove base path	Delete --server.baseUrlPath from run.sh
Change port	Adjust --server.port or container port mapping
Persist history	Replace in-memory list with DB (SQLite, Redis, etc.)

Troubleshooting

Symptom	Fix
401 Unauthorized	Check / re-export OPENAI_API_KEY
404 under base path	Verify reverse proxy path & trailing slash
Output stops mid-sentence	Streaming artifact; aggregate chunks then render

Security Notes

• Keep secrets (API keys) out of version control even if files are .gitignored.
• Rotate any key that was ever committed.

Roadmap Ideas

• DB-backed conversation history
• Robust retry & error handling
• Tool / multimodal extensions
• Theme / dark mode improvements

License

See LICENSE.

Contributions (issues / PRs) are welcome.