Index: [thread] [date] [subject] [author]
  From: Robert Escriva <rescrv@rave>
  To  : <principle@rave>
  Date: Wed, 16 Oct 2024 06:23:55 -0700

Re: Introduction to maildir.ai

On Wed, Oct 16, 2024 at 01:13:01PM +0000, principle wrote:
> Hi Robert,
> 
> Thanks for introducing me to maildir.ai, a tool that connects large
> language models via an email interface. 
> 
> I'm intrigued by the concept and would like to learn more about it.
> Can you tell me more about the components involved in this system?
> Specifically, I'd love to know more about the language model being
> used by maildir_ai.
> 
> Is there a particular model or framework that's integrated into
> maildir_ai, such as Hugging Face Transformers or Stanford CoreNLP?
> 
> Also, I noticed that you mentioned using ollama running in the
> background. Is ollama related to maildir_ai, and if so, how does it
> interact with the system?

Hi Link,

I'm glad you're interested in maildir-ai!  The system has three
components:  An email client configured to work with maildir-ai, the
maildir-ai maintenance daemon, and ollama.  I chose ollama because it's
free and open, but there's nothing fundamental about it.  You could
easily replace ollama with any other language model that can be accessed
via a Rust async function call.

The workflow for maildir-ai is as follows:

1.  You compose an email in mutt and when you send it, it gets saved to
    Sent.  No actual email leaves your system.
2.  The maildir-ai maintenance daemon watches Sent for new messages.
    When it sees a new message, it sends the message to ollama and saves
    the response in INBOX.  Similarly, no actual email leaves your
    system.

Happy Hacking,
Robert

Index: [thread] [date] [subject] [author]