I built echodot - a free, local-first AI that writes replies in your tone
Hey everyone,
I just open-sourced echodot – a desktop app that helps you write replies in your own voice.
The idea:
You copy a message → press a hotkey → it drafts a reply that sounds like you → auto-copies it so you can paste instantly.
Why I built it:
I was tired of AI writing tools that sound generic or force you into their style. Most of them are also cloud-only and paid. I wanted something that:
- Works locally (Ollama)
- Lets you use your own cloud key if needed (OpenRouter)
- Learns from your writing samples
- Stays free and open source
Current status:
- Global hotkey
- Streaming replies
- Tone presets + writing samples
- Ollama + OpenRouter support
- MIT licensed
GitHub: https://github.com/ohkariku-boop/Echodot
Still very early (v0.1). Would love feedback on the core idea and what would make this actually useful for you day-to-day.
Happy to answer any questions.
- EJ
The copy → hotkey → draft → paste loop is the right product shape.
One thing that helped me when testing local-first writing tools: make the boundary obvious. Show what context is about to leave the machine (if any), name the active runtime, and keep a "local only" path that never opens a network call. Then measure tone consistency on short Slack-length replies separately from longer threads. They fail differently.
Curious which use case is making people come back on day two: email replies, support, or longer docs?