Using OpenAI's 01-Preview Model In Cursor & OpenAI Playground

Show Notes

Parker walks through using OpenAI's 01-Preview model inside Cursor and the OpenAI Playground, showing a practical workflow to test prompts, prep a repo for LLMs, and implement a calendar-delete feature with smart prompts.

Setup: Enable 01-Preview in Cursor

• In Cursor, open Settings (top-right) > Models > Add Model.
• Select 01-D mini 01-Preview (the 01-Preview model).
• You’ll hit limits quickly. You have two options:
  1. Set a new hard limit (this can cost about $0.40 per request). Use sparingly.
  2. Use an OpenAI API key from platform.openai.com and paste it into Cursor (verify to enable).

Actionable note:

• If you’re prototyping or testing, the API key route avoids per-request throttling/costs the hard-limit path incurs.

Code snippet (conceptual):

# Enable 01-Preview in Cursor
Settings -> Models -> Add Model -> 01-D mini 01-Preview

Prompt optimization in OpenAI Playground

• In Playground, switch to the Assistant area.
• Bring in your Cursor rules:
  • Copy Cursor rules from the Cursor directory (the prompt rules you use for Cursor).
  • Paste them into the System Instructions in Playground.
  • Click Create to generate a system prompt tailored for the 01-Preview model.
• Reopen the chat in Playground:
  • Go to the Beta chat flow, paste the modified system prompt, and add it as the system prompt.
• Prep a target repo with repo pack:
  • Use repo pack to optimize the entire repo or selected directories/files for LLMs.
  • It outputs a structured text file (file summary, tree, repo files, etc.) at the root.
  • If you don’t want to dive into docs, you can invoke it via Command-K in your environment and type repo pack to generate the file.

Code snippet (conceptual):

# In Playground
1) Copy Cursor rules from Cursor directory
2) Paste into System Instructions
3) Click Create -> use the generated system prompt

Repo prep workflow with repo pack

• Run repo pack to generate a summarized, LLM-friendly view of the repo (file tree, relevant files, etc.).
• It spits out a nice summary file at the repo root. Drag that file into Playground to guide the model on your codebase.
• Example workflow described:
  • Generate repo-summary.txt at root
  • Paste its contents into Playground to inform code tasks
  • Use Playground to iterate on changes and get back results quickly

Simple outline of the flow (conceptual):

$ repo-pack
# outputs: repo-summary.txt at project root

Concrete task: Delete a calendar event with recurrence handling

• Objective: Add the ability to delete a single calendar event, handling recurring events gracefully.
• UI/UX reference: Model after Google Calendar’s dialog (show options for single instance vs. all future occurrences).
• Files mentioned: a calendar-related dialog TSX file (SheetDialog.tsx or similar) with:
  • Recurrence check
  • Options: delete this instance, delete all, delete future events
• Prompts and linting:
  • Include linting errors you’ve encountered and enforce strict typing to surface dead code.
  • Build the prompt to instruct the assistant to implement the new functionality accordingly.
• Execution flow:
  • Copy and paste the prompt suite (including objective and lint cues) into Playground.
  • Run the model to generate the implementation snippets, then iterate.
  • Playground tip: you can compare results or iterate with the “O” (or related) quick actions to re-run smaller changes quickly.
• Typical turnaround: about 2–5 minutes depending on prompt size and repo scope.

Takeaway workflow:

• Use Cursor rules to shape prompts
• Use Playground for rapid iteration and testing
• Use repo pack to prep your repo for LLM-friendly prompts
• Iterate on a concrete feature (calendar delete), leveraging strict typing and lint feedback to refine

Quick tips and caveats

• 01-Preview limits can bite; prefer API key for a smoother workflow or use sparingly for testing.
• Copy and adapt Cursor rules rather than rewriting from scratch—consistency pays off.
• The Playground is great for rapid testing, prompt tuning, and validating integration prompts before writing code.
• Strict typing and linting help catch dead code early when you’re feeding code tasks to the model.

Takeaways

• Add 01-Preview in Cursor, but manage limits intelligently (hard limit vs API key).
• Move cursor prompts into Playground as a system prompt to experiment and refine without losing context.
• Use repo pack to produce a concise repo summary for LLMs, then feed that into Playground to guide prompt-based code changes.
• For real features (like calendar deletion with recurrence logic), build a concrete prompt around UX patterns (Google Calendar style dialogs) and validate with linted, strongly-typed prompts.

Links

• OpenAI API keys
• OpenAI Playground
• Repomix (LLM prep tool)
• Cursor
• Parker Rex on X