Foley
A voice to-do list that files, plans and asks first
130 followers
A voice to-do list that files, plans and asks first
130 followers
Say what needs doing. Foley turns it into a task with the project, date and reminder already set, and asks when it isn't sure instead of guessing. When plans change, say so and it moves the task. Ask it to plan your day, or what to work on first, and it answers from your list. Capture from your Apple Watch. The desktop app is free with no account, using your own OpenRouter key. Foley Pro runs the AI for you and syncs every device. On iPhone, Android, Mac, Windows and the web.
Free Options
Launch Team
ClarifyThe Full-Stack AI CRM that replaces 5 tools
Promoted
"asks when it isn't sure instead of guessing" is the right default and most voice products get it backwards because asking feels like friction and guessing feels like magic, until the guess is wrong and now you've got a task filed under the wrong project with a made-up date. we fight the same tradeoff on calls at Dial. where's the line for you on what counts as "not sure", confidence score under some threshold, or ambiguity between two specific candidates like two projects with similar names?
@galdayan Great question, and you're absolutely right, Gal. Essentially, the way that I found it was that, as I was doing some testing, it kept coming up with specific things it had decided that it just didn't make sense.
One of the tests that I was doing was when I wanted to add the task, but I didn't specifically provide it the time. This could be an all-day task, of course, but generally, tasks have a specific time. This was more or less at the time that Jev came out. Jev makes the decision on whether this is clear or whether we are missing information or not.
Obviously, there could be much more improvement that I can do, but on a general basis, I think it's pretty good as it is in terms of asking follow-up questions. It kind of becomes a dialogue between yourself and Foley.
@joseairosa makes sense to let a model judge ambiguity rather than hand-tune a confidence score, but that trades one failure mode for another. a model's sense of "clear enough" can flip between two phrasings of basically the same request. have you noticed Foley asking on one wording of a task and just guessing on a slightly reworded version of the same thing, or has it been consistent so far?
@galdayan it's a fair push. So far I have sensed it to be consistent, but consistency is hard to guarantee, models are inherently non-deterministic, which means that it's in the nature of the models to analyse each request separately and there is no guarantee that it'll have the same outcome every time. I support it with a good amount of good old'fanshioned code to attempt to give some of that consistency.
@galdayan There is a layer of memory, for example, if you want anything that sounds like an appointment to be added to a particular project. These are called rules that get fed into the agent for him to know if needs to execute this regardless of what he thinks.
Congrats on the launch. I like that it asks before filing — half my voice dumps are just "call them back" with zero context. Does the clarifying step happen on-device or after sync?
@mostafahosseini Thank you! If you try it, let me know what you think and if you found any issues. On the specific case, yes, this could be classified as ambiguous and ask for a followup.
Will grab TestFlight this week — the ambiguous follow-up is exactly what I need when a voice note is just "sort the invoices" with no folder named.