Voice notes sent over Telegram become Obsidian thought notes with transcript, cleaned argument, linked context and a draft; Hermes also tracks recipes.
Voice notes sent over Telegram become Obsidian thought notes with transcript, cleaned argument, linked context and a draft; Hermes also tracks recipes.
Mete runs Hermes, an open-source agent framework, on a Mac Mini and messages it through Telegram. He sends rambling voice notes while walking his dog. The agent transcribes them, files an Obsidian note with metadata and tags, keeps the raw transcript, and adds a trimmed version, linked context and a draft. For business ideas it also researches the idea. Recipes arrive as links, messages or photos, and the agent files them in a uniform format it created itself.
He had structure and raw material ready by the time he sat down, and wrote the article himself. He calls the system messy and says most people don't need it in its current shape. Inbox processing is still missing.
He pays $100/month for a ChatGPT plan to power it, plus $5/month for Obsidian Sync. He says models that run locally on 16GB RAM can't drive the agent well.
When I send a voice note to [Telegram chat], transcribe it and create a note in my Obsidian vault at [vault path] with the transcript, a cleaned-up version of the argument, links to related existing notes, and a short draft of [blog post or message]. Reply with the note title and path. Never delete or overwrite existing notes; create new ones only.
A product manager had Hermes produce a weekly competitor briefing; its skill rewrote itself four times and by week six surfaced patterns he had missed manually.
A Make scenario runs at 9 AM, finds a YouTuber's newest video, has Hugging Face summarize it, and appends the summary to a Google Doc.
An n8n workflow scrapes Reddit posts and comments via Bright Data, extracts concerns and themes with Gemini, and writes reports to Notion.