Deepgram Saga - The Voice OS for Developers

Deepgram

Control your dev workflow with your voice. End the mentally expensive context switching that takes up almost half your day by skipping the alt-tabbing, clicking, and typing. Just speak to deploy, document, or debug. Powered by MCP, Saga keeps you in flow.

13d ago

Replies

Best

Sharon Yeh

Deepgram

Maker

📌

Hey Product Hunt Community! 👋 Sharon here, PM at Deepgram. We’ve been building Saga for the past few months to solve a pain that’s personal for most of us in PM and Engineering: too many tools, too many tabs, too much friction. Most AI copilots still need help translating your intent into precise prompts. Other voice assistants are not well integrated into the typical dev stack. So we thought: what if you could speak your commands into workflows? That’s Saga. It listens, understands, and actually executes. We built it for developers who are already using Cursor, Windsurf, Jira, Slack, Linear (and more), and we’d love your feedback as we expand across modalities, operating systems and breadth of tasks. 🙌 We’d love to hear from you on: One thing you wish you could do just by speaking? What MCP action is your favorite? Thanks for checking out Saga! We’ll be around all day in the comments and our Discord channel: https://discord.com/invite/deepgram. —Sharon (and the Deepgram team)

Report

13d ago

Chan Bartz

In your video you say that the AI can take in vague instructions and turn into precise instructions. Models like o1 and o3 do that sort-of fine, but the question holds: can it be done in a truly useful way? Would appreciate so much some use cases and other examples to see how it works in your cool app

Report

12d ago

Sharon Yeh

Deepgram

Maker

@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!

Report

12d ago