I construct a IMAP synced 💌 E mail composer – Neighborhood

I construct a IMAP synced E mail composer.

The software routinely generates each optimistic and detrimental skilled e-mail responses by analyzing your inbox messages utilizing AI, permitting you to decide on and edit earlier than sending by means of your common e-mail consumer.

Hello,

I constructed an AI-powered e-mail composer that integrates along with your e-mail inbox by way of IMAP. The system:

Reads incoming emails out of your inbox
Makes use of an LLM (Giant Language Mannequin) to routinely compose two response choices:
Knowledgeable optimistic response
Knowledgeable detrimental response
You possibly can edit both response earlier than sending. As soon as despatched, the e-mail syncs along with your mailbox’s despatched folder by means of IMAP, making certain it seems in your common e-mail consumer (whether or not you employ webmail, Thunderbird, or different e-mail instruments).

I used Go4lage for the techstack. https://go4lage.com/

For the LLM API I used gemini in the course of the dev course of after which I switched to Mistral for manufacturing. (I can sizzling swap if wanted – to any suited LLM api together with self hosted stuff)

Observe: There’s a demo video obtainable, however please bear in mind it’s in German.

I didn’t perceive a phrase however I watched the demo and acquired the gist! By “sizzling swap” do you imply the LLM is configurable within the app? Or do you imply you would need to recompile your binary?

There isn’t a want for recompile if sizzling swap is carried out.

Usually you may swap all LLM API’s so long as you don’t use the precise options. Frankly, I by no means wanted any of them, all of them really feel like they’re simply there to vendor lock the inexperienced developer in. There’s a little bit of adaption wanted stuff like writing specifc variants of prompts or perhaps additionally swap tokenizers, for those who want them, however it’s not too unhealthy.

I’ve developed one other app. This time for desktop. Translation and textual content fixing. And sure, it really works with all LLM APIs I’m conscious of. In fact, it additionally works with DeepSeek, regardless that some folks consistently act as if that had been one way or the other particular.

(This one runs domestically with ollama. qwen2.5:14b efficiency is OKyish)

Did you construct desktop app utilizing Wails? Or one thing else?

I used this one, I’m not too joyful as a result of it wants 2-3 seconds to start out. However perhaps I’m utilizing it incorrect.

I simply requested the kde.org for a mentor and hopefully we are able to embrace this in kmail and as ktext within the subsequent kde launch.

That seems like an awesome software for streamlining e-mail responses! Utilizing AI to generate each optimistic and detrimental skilled replies can save loads of time whereas making certain a elegant tone. Integrating it with IMAP for seamless inbox and despatched folder syncing is a brilliant method.

How has the transition from Gemini to Mistral impacted efficiency or response high quality? Additionally, any plans to help multilingual e-mail composition?