Disclaimer
Is this plugin open source? Yes — MIT, GitHub - swayinfo/elton-ai: Voice and text AI assistant for Obsidian: talks with your vault, searches and writes notes, files them by your own templates, and answers from your Telegram bot. · GitHub
Is this plugin completely free? Yes for the plugin itself — MIT, no paid features and no upsell inside it. It does need access to a language model, which you provide: your own OpenRouter/OpenAI key, or a local model via Ollama or LM Studio. I also run a provider service of my own as one of the options, so I’m answering “yes” with that stated openly rather than quietly.
Is this plugin vibe-coded beyond the author’s ability to comprehend how it works? No. I use AI assistance while writing code, but I understand what is in the repository, I maintain it, and I answer bug reports.
Community Directory: Elton AI
What it does
Talking to it. Hold a key anywhere in Obsidian, say what you want, and it answers out loud. There’s a normal text chat as well, so the voice part is optional.
Finding things. Keyword search plus semantic search over the vault, so answers are built from your notes rather than from the model’s general knowledge.
Writing notes the way your vault writes them. Before creating a note it looks at what properties the neighbouring notes in that folder carry, reads the template, and links the new note to its parent MOC. It can also create real Kanban boards rather than a note with headings that resembles one.
Moving things around. Renaming and moving notes with all links repaired, creating folders, opening several notes at once, and tab handling.
Reaching other plugins. It can execute any command from the command palette, so whatever your other plugins expose as commands is available to it.
Telegram. An optional bot: you write to it from your phone by text or by voice message, and the same assistant answers from your computer. It responds to exactly one person, whoever sends the first /start.
Scheduled automations. A morning digest written into the daily note.
Interface language is English or Russian.
Privacy
There is a list of private folders that are never read, never searched, never indexed and never sent anywhere. Outside those, only the fragments relevant to a request are sent to whichever model you configured. With a local model nothing leaves the machine at all, and speech synthesis is local too (Piper). The one piece I have no local option for is speech recognition, so talking to it currently requires a cloud key for that step. The README lists every address the plugin can contact.
How it relates to the AI plugins that already exist
Smart Connections does semantic connections fully locally with a bundled embedding model and no keys, and does that job better than I do. Copilot is the closest neighbour and is more mature as a chat and vault-QA plugin. Text Generator is template-driven generation into the current note. The things I built that I couldn’t get from them are the push-to-talk voice loop, the Telegram bridge, and the note creation that follows the vault’s own templates instead of producing generic markdown.
Install
Community plugins → Browse → “Elton AI”. A setup guide opens on first run.
Bug reports and feature requests: github elton ai — please put them there rather than in this thread, so this stays readable.