oai.py is a python script that provides a command-line interface (CLI) for interacting with reasoning GPTs via the OpenAI API. It functions as a REPL (Read-Eval-Print Loop), allowing for continuous conversation with an AI model.

• REPL Interface: Engages in a continuous conversation until you exit.
• Server-Side Conversations: Keeps track of the conversation context on the server-side.
• Command-Line Flags: Customize the model, budget, and other settings.
• Streaming Support: Can stream responses from the API for faster interaction.
• One-Shot Mode: Used for single query then exit immediately.

To start the REPL, run the script from your terminal:

• Python 3
• An OpenAI API key set as an environment variable: export OPENAI_API_KEY='your-key'

1. Clone the repository
2. Install the required dependencies:

uv sync
uv run python oai.py --stream

pip install -r requirements.txt
python oai.py --stream

You can then type your prompts and press Enter. To exit, type /quit or use Ctrl-C.

You will be in a conversation with looping prompting with memorized history of previous prompts and responses.

/reset to clear out the history and start a new conversation.

• /quit: Exit the REPL.
• /reset: Start a new conversation, clearing the previous context.

For single interactions, you can use the pipe input to the script.

echo "What is the capital of France?" | uv run python oai.py
The capital of France is Paris.

% uv run python oai.py --one-shot
Enter your prompt: What is the capital of France?
The capital of France is Paris.

The script supports several command-line arguments to customize its behavior:

• --model NAME: Specify the reasoning model to use (default: gpt-5-mini).
• --system TEXT: Provide system instructions for the AI. (default: "Answer directly and concisely.").
• --budgets "A,B,...": A comma-separated list of response token budgets to try (default: "800,1600").
• --effort CHOICE: The effort level for the model, from "low", "medium", or "high" (default: "medium").
• --stream: Enable streaming for the response.
• --one-shot: Use the script for a single interaction.

• Python 3
• openai library

You can install the required library using pip:

uv sync 
  -or- 
pip install openai

You also need to have your OpenAI API key set as an environment variable:

export OPENAI_API_KEY='your-api-key'

endpoint-oai.py is a Python script that provides a REPL (Read-Eval-Print Loop) for chatting with any OpenAI-compatible API endpoint, such as LM Studio. It's designed for continuous conversation with session-only memory management, automatically compressing old messages into a running summary to avoid exceeding the context window. It also includes a /check command to verify the accuracy of the last AI response using high-end OpenAI API. The /check command is useful when you're prompting smaller, less capable AI models.

• OpenAI-Compatible Endpoint: Connects to any service that provides a /v1/chat/completions endpoint.
• Automatic Context Compression: When the conversation gets long, it automatically summarizes the oldest parts of the dialogue, allowing for very long conversations.
• Response Verification: Use the /check command to get a second opinion on the AI's last answer from an OpenAI model (e.g., gpt-5-mini).
• REPL Interface: For continuous conversation.
• Configurable: Key parameters can be configured via environment variables.

To start the REPL, run the script from your terminal:

python endpoint-oai.py

• /quit or /exit: Exit the REPL.
• /reset: Clear the conversation history.
• /model <name>: Switch to a different model.
• /system <text>: Change the base system prompt.
• /summary: View the current compressed summary of the conversation.
• /summarize: Force an update of the conversation summary.
• /check: Verify the last AI response for accuracy using the OpenAI API.
• /help: Show the list of available commands.

The script can be configured using the following environment variables:

For the LM Studio / local endpoint:

• LMSTUDIO_BASE_URL: The base URL of the endpoint (default: http://127.0.0.1:11435/v1).
• LMSTUDIO_MODEL: The default model name to use.
• LMSTUDIO_API_KEY: The API key if required (default: not-needed).
• LMSTUDIO_KEEP_TURNS: Number of recent turns to keep in full detail (default: 8).
• LMSTUDIO_MAX_CONTEXT_CHARS: Character limit to trigger compression (default: 12000).
• LMSTUDIO_SUMMARY_WORDS: Target word count for the summary (default: 120).

For the /check command:

• OPENAI_API_KEY: Your OpenAI API key (required to use /check).
• OAI_CHECK_MODEL: The model to use for verification (default: gpt-5-mini).
• OAI_CHECK_DEBUG: Set to 1, true, or yes to print debug information.