Linoy Bar-Gal

Running Claude Code solo on Sorti means I context-switch constantly between writing features, debugging, and reviewing , which is not what it was designed for. The parallel session management is the thing I didn't know to ask for. What I'm really curious about is the diff review UX: when you're reviewing changes from multiple agents at once, how does the app surface conflicts or dependencies...

The quiet headline here is the AI having your whole canvas in context without re-explanation. That's the problem I hit constantly, I spend the first few minutes of every AI session just re-establishing what I'm building and why. The local-first architecture is a big bet too. Was that a values call about user privacy, or did you run into specific problems with cloud sync during development?

Building an iOS app where the core loop depends on someone recording their own voice is a high trust bar, most people feel weird hearing themselves back. The "how you came across" feedback angle is the interesting part though. most speech tools grade delivery mechanics but miss the intent gap entirely. Did you build the AI feedback layer yourself, or is it layered on an existing API? Shipping...

AI that challenges assumptions, is much harder to ship than AI that confirms them and honestly the more useful one. As a solo founder I've spent weeks building confidently in a direction that turned out to be wrong, and there was nobody in the room to push back. Does Athena challenge you proactively mid-session, or is it more structured like a pre-spec review moment? Congrats on the launch.

The shared context across video, image, and audio is what makes this different. As a solo founder handling all my own social content, the biggest time sink isn't generating it's re-explaining the brand context every time I switch between tools. If this actually maintains creative direction across a full campaign pipeline, that's a real unlock for small teams. Congrats on the launch.

This solves a problem I didn't realize I had until I saw it. I've been building with Claude Code daily and my skills folder is a mess duplicates, things I set up once and forgot about, broken ones I never cleaned up. The audit plus dedup combo is exactly right. Love that it's open source too. What was the hardest part detecting overlapping skills that use different wording?

"Not a dashboard, a co-worker" this positioning resonates. As a solo founder, I don't have a data team to investigate why a metric dropped. I just see the drop and panic. The idea of an AI that can connect the dots between a code release and a metric change is exactly what small teams need. How granular does the root cause analysis get?

98% token reduction is wild. I've been working with MCP servers and the tool list bloat is a real problem the agent spends half its context just reading tool descriptions. The zero-config approach is smart because the moment you ask devs to configure which tools to expose, you've already lost. How does it decide which tools to surface for a given query?

This is the kind of open-source release that actually moves the ecosystem forward. I'm building a consumer app solo and the idea of background agents handling PRs while I focus on product decisions is the dream. Really curious about the sandbox orchestration layer, how do you handle state when an agent needs context from multiple repos? Excited to dig into the source.

"No new app to learn" - this is such an underrated product decision. As a solo founder drowning in email, the idea of just messaging an assistant on WhatsApp instead of opening another dashboard is really appealing. How are you handling the trust barrier? Giving an AI access to your inbox feels like a big ask, even when you desperately need the help.

The "design context for agents" gap is real I keep watching Claude and Cursor hallucinate component names and tokens because they just don't know what my design system actually looks like. What does the MCP expose today: tokens, components, Figma layers, usage examples? And does it update live as the design system changes, or is it a snapshot?

Custom keyboards are one of the most underrated hard problems on iOS, App Extensions, memory limits, the whole "Full Access" trust dance with users. Nicely done shipping this. Quick one: are you running the model fully on-device, or is this a hybrid with a server hop for heavier rewrites? The latency tradeoff there is brutal and I'm curious how you landed.

35.6% is an oddly specific number, which makes me trust it more than "save up to 50%." What's actually being compressed, prompt-side context pruning, response caching, or something closer to semantic dedup across a session? Asking because I've been eyeing my own API bill lately and the honest breakdown matters.

A terminal CFO is such a niche-but-obvious idea the kind of tool you only build because you personally needed it. Curious about the storage layer: are you keeping financial data fully local, or is there a sync component? And how are you handling category inference for transactions LLM every time, or a cached classifier? Starring the repo.

This hits a nerve. As a solo founder, I have this spreadsheet that's supposed to track what I'm paying for OpenAI, Vercel, various APIs, and it's permanently out of date. The moment you forget to log one new tool, the whole system breaks. Does Orbit auto-detect subscriptions from bank transactions, or does it hook into the billing APIs of each service directly? The second approach sounds way...

"For people who want less" is such a bold tagline in a space where every product promises to help you do more. I respect that. There's something broken about the default assumption that productivity means throughput. Curious what the hardest design decision was, I'd guess it's knowing what to leave out, because every feature you cut means someone emails you asking for it.

The MCP integration is what caught my eye, I use MCP in my own stack and the idea of it feeding directly into structured reports instead of me copy-pasting across 5 tools is exactly what I need. How opinionated is the output format, or can you customize it?

"AI still feels fake", yes. I've been trying to build content around my app launch and the gap between how I talk naturally vs. what comes out when I try to "create content" is painful. Does this work with solo recordings (Loom-style) or only multi-person meetings?

The Agent Skills concept is the interesting part here, turning runbooks into reusable automations instead of just chatting with your terminal. How do you scope execution permissions? if I point this at a production cluster, what's the boundary between "suggest a fix" and "apply a fix"?

The Metal shaders are gorgeous. Honest question though- how do these perform on older hardware? I'm thinking iPhone 12 era and below where the GPU budget gets tight fast. Also curious if you hit any SwiftUI layout thrashing issues when composing multiple animations on one view.