Bevel AI prescribed that I run 400 m x 6, with rests between each internal. After the run, Bevel knows only my overall average pace which includes the walking rests. Why doesn’t Bevel know my internal paces even though Apple does? — I used WorkOutDoors for the run and the screenshots show what was written to Apple Health.
Food can be logged with any of these options, and I use all of them depending on the situation. Just wanting to get confirmation from the devs that all the ones circled in blue use either the food data base or the old AI, and that only “Ask Bevel” circled in yellow logs food using the new Bevel Intelligence which is more accurate than the old “Describe Food” or “Import Food” or “Capture Food”?
I’ve been using Beville for almost 2 years and I’m accustomed to this happening when I travel from the US to Europe, especially with the sleep data that gets thrown off during the flights and the adding and subtracting eight hours from travel days, but this just happened by traveling just one time zone away in the United States yesterday. It loses all my historic data, temporarily. When I get back into my home time zone next week I’m sure it’ll all be back. It may be back even before then, but it’s very common for Beth to struggle with time zone changes in regards to maintaining historic data.
Edit: The solution per the devs is to Clear Cache and Re-sync Data. That solved it.
The default for the rest timer is to start at zero and go UP, buzzing every minute. If I open my phone and open the rest timer I can have it count DOWN from whatever time is set, but I have to have my phone with me to do that.
Is there a way on the Apple Watch Bevel App to have the rest timer count DOWN rather than up?
When I dictate messages to BI, the transcription often truncates after about 10 lines of text or 45 seconds. And it’s undiscoverable until after I finish speaking, click send, and to look down and see much of what I just said chopped off with an ellipsis…. ChatGPT handles over 6 minute rambles, and Claude and Grok have about the same limits. I don’t expect BI to listen for 6 minutes, but can we up the limit to 2 minutes or to 30 lines of text?
I’m not sure if this is new or not as I had not been entering food the AI previously. But I discovered this over the last couple of days that when I enter food via the AI that it defaults to log it at the date and time that the conversation started which in some cases was yesterday so I found that my food was being entered on yesterday‘s date and not today’s. It’s an easy manual fix for me to do. I just have to recognize that it was happening and then manually change the date and time of the entry when I verify it
Disclosure up front: I have no financial relationship with Bevel. No referral benefits, no compensation of any kind. Just a user.
I’ve been a user of Bevel Health since almost the very beginning. I’m also a heavy user of AI generally, specifically ChatGPT, Claude AI, and Grok, and I’ve spent a significant amount of time with all three, learning how they work, their limitations, and their parameters. Specifically, I’ve tried to use all of them as health and fitness coaches, with more or less (usually less) success. Grok tends to have pretty solid up-to-date internet access. Claude can connect to your Google Drive and Google Calendar if you give it permission. I’ve experimented with storing health and training data on external sites and feeding it to these AIs just to get something useful back in a coaching context. Using them as motivators, as secondary brains, as memory storage, as accountability sources, as cheerleaders, whatever you want to call it. It works, but it requires workarounds, and that’s the key challenge.
I didn’t get a whole lot of benefit from the AI in Bevel 2.0. But I’ve been using 3.0 for the last few days now, and I have to say that every single time I use it, every single session, I discover something new that impresses me. I posted a couple of times in the last few days here about it, but having used it again today I just want to say plainly that it is absolutely the best. It is not just a great fitness app. It is also probably the best AI implementations I have encountered, period.
Certainly, Bevel is using one of the major LLMs as its source, whether that’s OpenAI, Anthropic, or xAI. That’s not really relevant. Bevel isn’t building its own AI from scratch. It’s licensing one of those, and management has essentially said as much. The benefit, and this is the key point, is in how well it integrates everything. Bevel is a health app first and an AI second. Because of that, the AI has native, primary access to all of the data. It is innately tracking my sleep, my nutrition, my exercise, my recovery, my heart rate, my health data, the medical information I’ve entered, my training plans. Everything. And then it layers the high quality AI capabilities just under all of that. It has primary access to all of that information from the start, so I don’t need to find a workaround to give it context. And it’s built within its hierarchy to function as a coach first, while still being able to do everything else you’d want from a modern AI, in terms of providing motivation, accountability, and coaching support.
I’m also extremely impressed with how Bevel manages memory. It had a folder structure that makes everything very accessible. Some of the other AI platforms give you limited visibility into what they’ve stored about you, don’t structure it particularly well, and make it genuinely difficult to find or modify. Bevel makes the memory easy to get to and to modify, which means the AI can actually become exactly what you need it to be over time.
I’m sure that as I continue using it I will find things I’d like to see improved, and when I do I’ll either submit them through the suggestion feature in the app or post about them here occasionally. But what keeps surprising me, pleasantly, is how responsive this company is to that feedback. For a company of this size to maintain that level of responsiveness is genuinely rare, and it continues to impress me. This app is worth every bit that people pay for it. The more I discover, the more kudos I have for the Bevel Health team. It just keeps getting better.
If there is one big request I would make of the Bevel AI at this point, and I recognize it is a big one and I’m sure there is real cost involved, it would be a quality speaking voice. Something along the lines of what Grok or ElevenLabs offers. A significant amount of the communication processing I do is while I’m driving, and while I’m able to dictate to Bevel AI just fine, what I can’t do is tap a play button and have it read the response back to me the way I can with Claude, Grok, or ChatGPT. That functionality doesn’t currently exist in Bevel, and it’s something I would genuinely pay extra for. My current workaround is copying the response from Bevel, pasting it into Grok, and having Grok read it to me, which is more taps than anyone should be doing behind the wheel. It would be far better to just have Bevel speak for itself. I know this one is probably not coming anytime soon, but I’ll say this: the day Bevel adds a quality voice output, I will be canceling my subscriptions to every other AI I currently use. That’s how much I think of where this app is headed.
Im absolutely loving the new AI but I don’t like that it has a list of questions after each response it gives. Too often I accidentally tap on one and it derails the conversation. I’d like there to be a way to turn off those suggested follow up questions.
What folders does everyone have in their AI? These were created by the AI, not me. Just curious if anyone has anything different.
My folders are…
Core
—Personally
—User
Memories
—Habits
—Likes
—Dislikes
—Goals
—Nutrition
—Events
Loving the 3.0 update, but food recognition and input still need more improvement.
The announcement video showed Grey taking a photo of his meal, so I hoped food recognition would be better in 3.0 because of that. (My experiences with photo logging in V2 were pretty underwhelming.)
Yesterday, I took a photo of my breakfast buffet plate with about six items on it, including a cube of pâté. The app correctly identified every item, including the pâté, but it showed the pâté as zero calories. When I clicked on it, all the nutritional information fields were blank or zeroed out. I had to delete that entry and manually search for a new “pâté” entry, then pick the first reasonable-looking option. That’s Issue #1.
Issue #2 is serving size estimation. Historically, V2 often overestimated portions or defaulted to a full serving or whole item, which might be 5x more than what’s actually in the photo. More often than not, I had to manually change the serving size from 1.0 down to 0.4 or whatever I estimated it to be.
Issue #3 is finding foods I’ve logged before. If I start typing something I’ve eaten dozens of times over the last several months, for example “egg salad sandwich,” I think it would prioritize the exact item I’ve previously logged. Instead, when I type “egg sal…,” I get generic results like two versions of “Egg Salad,” two versions of “Egg,” and “Eggcellent Egg Salad” (which I’ve never eaten nor even heard of).
I tap “More” and scroll, but the food I’ve logged before still doesn’t appear. Then I remember it was originally entered as “Southwest Egg Salad Sandwich,” so I start typing “South…” and again get a bunch of irrelevant options. Still nothing I’ve actually eaten. Only after remembering that I need to tap “My Foods,” and then tap “Historical,” do I finally find it. That’s way too much work for a food I eat fairly regularly, just not often enough for me to make it into a custom food.
Issue #4, which ties into #3, is how search works inside My Foods → Historical. If I type “Sou…” looking for that Southwest Egg Salad Sandwich, it pulls up every food with “sou” anywhere in the title and sorts them alphabetically. So the first result is “Appetizers Baskets Garden Fresh Salads Hearty Soup” from 18 months ago. It’s first on the list because it starts with “A.” Next are the Cs: chicken noodle soups from the last two years and some old cream of corn soup. I have to scroll all the way down to “S” to find my Southwest Egg Salad Sandwich that I ate last week, and the week prior. Historical foods should be sortable by recency, most recent first, not alphabetically.
I’m looking for a version of “American Pie” sung by Don McLean where the song is repeatedly paused and a male voice comes in to explain the lyrics, essentially giving the interpretation of each section. I heard it twice on the radio in the U.S., sometime in the 1980s or 1990s. Can’t find it now.