I built a desktop app AI harness that keeps local meeting notes and can use them to autocomplete my typing as well as complement my dictation with additional context via reading my screen or using the previous meetings notes.
It's been a bit of a struggle to get it to work well and also to not be too intrusive and laggy. I've finally made it usable in the last few days so I'm sharing it here. I've done a lot of experiments with different models and quants and Gemma 4 E2B base at 6bit is the best balance of quality suggestions and low latency on a regular MacBook.
What's different from Cotypist and Cotabby
License / price — Cotypist: Proprietary, free for 100 words a day, $$ for unlimited; Cotabby: AGPL-3.0, free (beta); LokalBot: GPLv3, free, no limits
Model — Cotypist: Local (Gemma 4 E4B recommended); Cotabby: Local GGUF via llama.cpp, or Apple Intelligence; LokalBot: Gemma 4 E2B base Q6_K, in-process llama.cpp
Knows what was said in your meetings — Cotypist: No; Cotabby: No; LokalBot: Yes, opt-in: up to two facts from your own notes, when your draft names something distinctive
Text above the field — Cotypist: Not documented on its site; Cotabby: OCR of the window; LokalBot: Accessibility text above the field, on by default, never from private windows
Also in the app — Cotypist: Autocomplete; Cotabby: Emoji, macros, autocorrect; LokalBot: Meetings, dictation, screen recall, sharing one model RAM budget
If you try it, I'd love to hear how it writes for you: which app, which language, and what it suggested. And if you know a small base model that beats Gemma 4 E2B at about 50 ms, I'll benchmark it.
1 comment
[ 5.2 ms ] story [ 15.8 ms ] threadIt's been a bit of a struggle to get it to work well and also to not be too intrusive and laggy. I've finally made it usable in the last few days so I'm sharing it here. I've done a lot of experiments with different models and quants and Gemma 4 E2B base at 6bit is the best balance of quality suggestions and low latency on a regular MacBook.
What's different from Cotypist and Cotabby
License / price — Cotypist: Proprietary, free for 100 words a day, $$ for unlimited; Cotabby: AGPL-3.0, free (beta); LokalBot: GPLv3, free, no limits
Model — Cotypist: Local (Gemma 4 E4B recommended); Cotabby: Local GGUF via llama.cpp, or Apple Intelligence; LokalBot: Gemma 4 E2B base Q6_K, in-process llama.cpp
Knows what was said in your meetings — Cotypist: No; Cotabby: No; LokalBot: Yes, opt-in: up to two facts from your own notes, when your draft names something distinctive
Text above the field — Cotypist: Not documented on its site; Cotabby: OCR of the window; LokalBot: Accessibility text above the field, on by default, never from private windows
Also in the app — Cotypist: Autocomplete; Cotabby: Emoji, macros, autocorrect; LokalBot: Meetings, dictation, screen recall, sharing one model RAM budget
GitHub (GPLv3): https://github.com/stevyhacker/lokalbot
If you try it, I'd love to hear how it writes for you: which app, which language, and what it suggested. And if you know a small base model that beats Gemma 4 E2B at about 50 ms, I'll benchmark it.