Openbase

This is the hard part and you're putting it front and center: which calls the agent makes alone vs. which ones it asks about out loud.We ship agents in a very different domain — they message customers and take payments for small businesses — and the rule we landed on after getting it wrong is: plan freely, execute visibly, but always ask before anything that touches money, anything that goes out to many people at once, and anything that changes configuration.The trap we hit: we tried letting the agent tune its own thresholds, with a guard scoring the outcomes. It optimized straight into the failure mode — a safe, wrong answer scores beautifully, because it generates no escalation and burns fewer tokens. We killed the autonomy and kept the human in the loop.Curious how you think about that when the approval is by voice, where "yes" is a very cheap thing to say.

I'm non-technical and ship almost everything through coding agents, so the part that would decide it for me isn't approval, it's verification. My recurring failure isn't an agent doing something destructive, it's an agent telling me a change shipped when it actually landed on a branch that had already been merged. Does Openbase report back from the actual repo state after the fact, or only from what the agent says it did?

the safety word answer above covers accidental triggers, but what about a garbled connection mid-approval, like walking through a spot with bad signal while saying yes to a PR merge? if the transcription comes back partial or mangled, does it fail closed and ask you to repeat, or is there any path where a mis-heard word could get treated as an approval for something you didn't actually confirm

Honestly the voice steering thing is kind of wild, I was telling an agent to fix a bug while walking the dog and it just committed the patch. Felt like cheating honestly.

Good to know it fails safe rather than proceeding on a guess - that queue-and-resume behavior is exactly what I'd want if I'm the one walking around unable to babysit the connection. Thanks for the detail.

Great idea, it’s the worst when you’re running an agent process and step away, screen goes dark and all work pauses. Best of luck!

Update: we've also just uploaded our YouTube demo to our launch, check it out here: https://www.youtube.com/watch?v=SsnUgItdae4

Congrats Lucas! I do almost everything by voice, my dictation is full of half-sentences and “wait, no, actually” corrections, so this launch is personally relevant. My question is about the input side rather than the approval side everyone’s covering: when my spoken instruction is messy or ambiguous, does OpenBase confirm its interpretation back to me before dispatching the agent? A five-second “here’s what I understood, go?” would save a twenty-minute run building the wrong thing. Curious how much you’ve tuned for how people actually talk versus how demos talk.

this solves a real problem, I live in Claude Code all day too and the "still tied to a laptop to babysit it" gap is exactly right. question that's more about reliability than the approval-safety angle everyone's asking about: what happens if you lose connectivity mid-flow, say you're walking and your phone drops signal right as an agent is waiting on a voice approval. does it just pause and wait indefinitely for you to reconnect, timeout and roll back the pending action, or keep going with whatever the last confirmed state was?

Sign me up, this is actually a dream! Hopefully I can find time to download this to try it. I like it because I used to drive around calling my human assistants to help me do work as I was driving long distances. This is really useful for people who drive long distances and are business people or people with work that requires computers. Maybe even artists can use this like music industry people.

Any plans to let teams create reusable voice workflows for common development tasks? That could save a lot of repeated instructions.

I usually have coding agents running while I’m working on something else, and sometimes I only need to approve a command, correct the direction or quickly check what changed. Being able to do that from my phone instead of going back to the desk every time is actually a good idea. Congrats on the launch!!

The idea of kicking off a feature while walking and coming back to a ready PR sounds both extremely useful and slightly dangerous :))

Love the idea! I even hoped for something quite similar about a week and a half ago while at work. The 'no screen, no desk' part hit me harder than you probably intended, because I'm not a developer.I work in a bakery. Factory rules mean my phone stays in the locker room, and my hands are busy for the whole shift. In the week before I launched my own app, some of my best thinking happened while I was working, ideas, prompts, things I wanted to try, all of it fully formed in my head and completely stranded until I clocked out or took a break. I kept thinking how good it would be to just talk it into a headset and have the work already moving by the time I got home.The honest reason I haven't tried agents yet is different from the reason you're solving for. It isn't that supervising pulls me back to a laptop. It's that I don't understand them well enough to trust one unsupervised, and I'm afraid of something getting deleted or overwritten without me noticing until much later. So my question is about the approval layer. When an agent is working and I'm not looking at a screen, what actually stops a destructive change from going through, and how much does it explain out loud before it commits? For someone like me, that answer matters more than the speed.

Congratulations on the launch, Lucas. Voice-based supervision is an interesting answer to the problem of agents working asynchronously while humans remain tied to a screen. How do you handle approval boundaries for sensitive actions such as commits, pushes, and pull requests?