Feature Ideas

The transcript quality is not good at all now. Suggest to focus on it and improve the transcript quality. When you launch a new transcript model, you can run on previous voice recordings and further improve the historical transcripts too.

Allow us to edit the transcription to improve the quality of the generated meeting summary. Separate request to edit Live Notes: https://feedback.limitless.ai/b/kvj1nlve/feature-ideas/edit-notes-created-by-limitless

Allowing the resources available on the Limitless app to be consumed by other AI products fits well into the AI agentic ecosystem.

The ability to search the transcript by speaker name, date/time, location, keyword ...etc would be great for quickly making the transcript useful without having to necessarily go through the AI model.

Native Android app

We have existing meeting audio or voice recordings that we want to transcript to add to the knowledge database.

Right now, it only captures the words. There's no indication of the sentiment or emotion. For example, I had it transcribe while I was playing a YouTube video of an SNL skit. It was about a doctor informing the family that their father didn't survive the operation. It was all humor and silliness. But from the transcript, you'd think it was a serious conversation. If Limitless could pick on on things like laughter and tone of voice to get a sense of the humor or emotion behind the words and somehow note that in the transcript, that would be awesome!

I shouldn't need to manually label 20 seconds for each new voice—Limitless should be able to patch together similar voices and say "Hey, we recognized a new voice... who's this?"

Have agents take actions on your behalf using the context from Limitless.

Sometimes using system mic is not an option (when I'm in a public space) so I'd like to be able to record both sides of the conversation on my Windows and mobile devices.

Where is often as important as when and what was heard. I want to browse the timeline of my life with location, audio and video.

Show recordings made on the Web/Mac app in the lifelog.

Allow users to pick their own LLM source. Right now, I use a mixture of Groq (because it's fast and has a strict no-training privacy policy), and smaller localized LLMs that I run through LM Studio or Ollama. I'd only really want to purchase and use Limitless if there were an option to do the processing of the material locally, especially since my use-case would be everyday life far more than it would be for meetings in a work environment. I want to have a record of the really awesome bit of philosophy my wife just told me about while writing her book, but I don't want that processed outside of my home. I've seen some other calls for this alluded to, but none that actually use the specific language I'm hoping to see implemented. It's just a matter of choice. Someone who wants total plug and play could just pay to have Limitless provide unfettered access to an OpenAI model. But this really should come with the option to pick your own OpenAI compatible API keys, and also the ability to connect local models. I LOVE the End-to-End Encryption that will be built into this, but e2ee isn't, alone, proof of material being leaked. I'm still not clear if the e2ee plans included the user controlling their own keys (but from some of the other comments here it doesn't seem like it?). Regardless, e2ee paired with a local-first mentality would be key. Unless Limitless can somehow go truly zero-knowledge, but ZK LLMs are still a pretty complex subject. As an early tech adopter, I'm focusing on devices that prioritize privacy and security, and that's what led me to Limitless in the first place.

Some speaker identities in the transcription are inconsistent. For example, Speaker A is now labeled as both Speaker A and Speaker C. This likely happened because both speakers have similar voices and were mistakenly identified as the same person, especially when they talked over each other. When I try to change Speaker C back to Speaker A, it updates all instances throughout the transcription, treating Speakers A and C as the same person. Unfortunately, there doesn't seem to be a way to reset them to ensure they are identified correctly again.

Being able to support other types of calendars and email clients like Outlook.