[MlMt] Local Email Client & AI Agents

[Robert M. Münch]

01-Jun-2026 -- 1:12 -- Quinn Comendant via mailmate:

> I agree with the others that it would be out-of-scope for MailMate to have integrated AI features.

Hi, sure, that's why my subject reads "&" and not "integrate".

> I think the correct feature boundary is for MM to remain a high quality IMAP client with *flexible integration with third party applications* – which is exactly what the [product description](https://freron.com/) says.

Yep, that was the idea: to start the discussion, not to integrate AI directly. For what should that be useful?

> MM's APIs already allow me to use AI with MM with custom bundles, e.g., I have an `Add to Reminders` command that takes selected message(s), runs them through a local LLM to summarize, then creates a Reminders.app item with the summary and URL linked back to the email.

The bundle way (I haven't deeply dug into it yet, so I might be wrong) looks to be human-triggered-centric. However, with agents, we have two sides triggering actions: humans and agents.

> One of the pain points for me now is finding specific information in a cluster of email threads. For example, I need to find a design detail mentioned in 100s of emails regarding a client project. This can be solved by using semantic/NL/vector search or dumping all the messages into the context window of a LLM.

That was my point regarding MM's backend storage, in case it would be a good AI agent access point.

> Currently, my workflow is to export `.eml` files to a folder, then use Claude to find what I'm looking for. It's cumbersome because raw `.eml` messages can't directly be searched (maybe some are base64 encoded), so Claude has to write a script to process the messages.

Thanks for that insight, that's exactly what I had in mind. And I see this as the "blocking point" of integration.

Of course, one could make a copy, try to keep a synced dataset, or work directly against the Gmail interface. However, at some point, I need to display the AI agent's results in the human interface. And most likely, it makes sense to identify them.

> I've tried hybrid search tools like Antfly or Vera, but they have the same problem of being unable to read email files directly. There are mail-specific search tools, but I haven't found any that use a semantic/NL/vector search index. It would help if MM was able to export messages as plain text, but that's not an ultimate solution.

I think these hybrid search tools are a superset of MM's search capabilities today. And making the storage backend a simple-to-use access point allows many new use cases. MM is the IMAP manager and provides a nice user interface, while I can still have some backend tasks work with the data.

Or, if MM becomes an MCP server, provides an API, or whatever, any agent can work with this interface.

> So, I'm not sure how to integrate this functionality to MM. I wouldn't want to add LLM/semantic search to MM, but I do need a way for LLM/semantic search tools to access the mail. Any ideas?

I would love to have semantic search integrated, which would take MM to a completely new level. And users are free not to use it. Mail is still the primary communication channel, and we now have much better ways to handle information encoded in plain language. To me, it's a very natural fit.

Viele Grüsse

--

Robert M. Münch
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.freron.com/pipermail/mailmate/attachments/20260601/c42d4ea8/attachment-0001.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 4076 bytes
Desc: S/MIME digital signature
URL: <http://lists.freron.com/pipermail/mailmate/attachments/20260601/c42d4ea8/attachment-0001.bin>