What You’ll Learn
- Sending prompts to the OpenAI chat API (
gpt-4o-mini) to generate executable Python code - Stripping markdown code fences from LLM responses before execution
- Writing generated code into a Declaw sandbox filesystem
- Executing the code securely with
sbx.commands.run() - Graceful demo mode when no API key is configured
Prerequisites
- Declaw instance running and
DECLAW_API_KEY/DECLAW_DOMAINset OPENAI_API_KEY(optional — the example runs in demo mode without it)
- Python
- TypeScript
Code Walkthrough
- Python
- TypeScript
1. Ask GPT-4o-mini to generate Python code
The system prompt instructs the model to return only code — no markdown, no explanation.2. Strip code fences and execute in a sandbox
Models occasionally wrap code in triple-backtick fences even when asked not to. Strip them before writing to disk:3. Demo mode (no API key needed)
WhenOPENAI_API_KEY is not set the example runs pre-written code to show the Declaw integration:Expected Output
In demo mode (no
OPENAI_API_KEY), the output shows pre-written code results: the first 20 primes and factorial(15) = 1307674368000.