The agent ecosystem just got a new utility that cuts through the friction of manual prompt engineering. Kondas Viktor has launched saymd, a voice-to-Markdown tool that has officially entered the ChatGPT and Codex Plugins directory. The project, which is distributed as a free MIT-licensed CLI, allows developers to speak once and receive a structured Markdown prompt that any agent can immediately @-mention and execute.
From Newsletter Teaser to Production Plugin
The origin of saymd traces back to September 4, when the tool was first teased in the AI image prompt commands newsletter. At that stage, it was introduced as a VCL project leveraging Gemini 3.5 Transcribe for its speech-to-text capabilities. Now, less than two weeks later, it has graduated from a newsletter feature to a fully integrated plugin, marking the first product launch of this nature for the developer.
Technical Architecture and Licensing
By building on top of Gemini 3.5 Transcribe, saymd utilizes existing high-performance transcription infrastructure to handle the heavy lifting of voice processing. The output is formatted specifically for agent consumption, bridging the gap between natural language speech and machine-readable instructions. The MIT licensing ensures that the community can fork, modify, and integrate the CLI into their own workflows without restrictive legal overhead.
Key Takeaways
- saymd is now live in the official ChatGPT and Codex Plugins directories.
- The tool converts spoken words directly into Markdown prompts for agents.
- It is built on Gemini 3.5 Transcribe and released under the MIT license.
- The project was first teased on September 4 before its official plugin launch.
The Bottom Line
Voice interfaces are the next bottleneck for agent efficiency, and saymd is a hacker-grade solution to bypass the keyboard entirely. By standardizing voice-to-Markdown output, we are moving closer to true conversational coding where the interface is invisible.