Deepgram

@chan_bartz Great question! Yes, models like o1 and o3 can kind of handle vague input, but they’re inconsistent without the right prompt structure. What Saga does is convert your fuzzy, natural speech into a clean, structured instruction that actually works when passed into tools like Cursor, Replit, or Windsurf. It acts like a pre-processor that speaks “LLM,” so you don’t have to.

Example 1:

You say: “Make a helper to format a date string”

Saga rewrites it into:

Create a helper function to format a date string. The function should:

1. Accept a date string as input.

2. Convert the date string into a more readable format (e.g., "YYYY-MM-DD" to "Month Day, Year").

3. Handle invalid date inputs gracefully by returning an error message or null.

Then pipes that into Cursor and gets back usable code on the first try.

Example 2:

You say: “Add error handling to this function”

Saga rewrites it into something like:

Add error handling to the specified function. Ensure that you include:

1. Try-catch blocks to manage exceptions.

2. Clear error messages for debugging purposes.

3. Return appropriate responses or throw errors based on the situation.

We’re seeing devs use it to avoid prompt tinkering and get more consistent results from AI coding tools.

Would love to see how it works for you when you try it out!