This week's AI tool launches: a few genuinely interesting ideas, a lot of Futurepedia noise

Every week I run StackIndex AI's discovery engine across Product Hunt and Futurepedia, dedupe the results, and write up what actually caught my attention. This week had a decent signal-to-noise ratio on the PH side. Futurepedia, less so — I'll explain why.

The Futurepedia caveat, upfront

Futurepedia surfaces tools when their listing gets edited, not only when something genuinely launches. So several of the tools that showed up in our feed this week — Revuze, Proscia, Kognitos — are established products that someone probably updated a description or added a screenshot. They're not new. I'd rather tell you that than pretend we caught eight fresh launches when we didn't.

The Futurepedia entries worth a glance this week:

• Dili — certified payroll compliance for construction projects with prevailing wage rules. Narrow vertical, but that narrowness is actually the point. If you're in that world, generic tools don't cut it.
• TradeOS — market analysis via MCP-connected AI clients. MCP keeps showing up. Worth noting.
• LiaPlus AI — voice agents for routine customer support calls. Crowded space, but voice agent quality is finally getting usable.

The rest I'd file under "tools that exist."

Product Hunt: the more interesting half

NoMac is the one I kept thinking about. The pitch: a Mac your AI agent can use, billed per minute, no subscription. Practically, this means you're renting a macOS environment that an agent can control — running apps, clicking around, doing things that need a real desktop OS rather than a headless server. Per-minute billing is the right model for this use case. You don't want to pay a monthly fee for something you spin up occasionally to run a workflow. Whether the execution is solid I can't say from a listing, but the pricing logic makes sense.

AXL describes itself as a GTM copilot for AI product founders with 1,200+ actions via MCP. I'm mildly skeptical of "1,200+ actions" as a headline number — that's the kind of figure that can mean anything — but the target audience is specific enough that it's worth watching. Solo founders building AI products don't have a sales team. Anything that handles pipeline work without requiring a CRM administrator is at least interesting in theory.

Agent Clan lets you control Claude Code and Codex agents with gestures. It's a weird one. Gesture-based interfaces for coding agents is a narrow enough use case that I genuinely can't tell if this is a clever ergonomics improvement for people running multiple agents, or a solution to a problem nobody has. I'll be curious to see retention numbers if they ever share them.

Sage2 by Levanto Labs turns live conversation into structured data and actions. Real-time transcription into structured output is a well-trodden path, but the "instant actions" part — triggering something downstream from what was said — is where the actual value would be. Depends entirely on what integrations they've built.

Inkling 2 Small is billed as a model that can see and refine what it builds. Vision-in-the-loop for code/UI generation. The "Small" in the name suggests they're positioning this as a lightweight option, which is smart if the quality holds — not every task needs a frontier model.

Frameboy by World Labs is interactive world exploration with something called Atlas. World Labs is Fei-Fei Li's company, so there's real research backing here. The product description is vague but the underlying tech — spatial intelligence, 3D scene understanding — is genuinely frontier work. Worth keeping an eye on even if the current product is early.

Memorias is an AI biographer that turns your stories into a book. Consumer-facing, clearly. Not my use case, but the market for this (older adults, families wanting to preserve history) is real and underserved by tech generally.

What I took away

The MCP pattern keeps appearing — TradeOS on the Futurepedia side, AXL on the PH side. That protocol is becoming infrastructure fast, and tools that assume it rather than explaining it suggest the developer audience is already past the "what is MCP" stage.

Per-minute or usage-based pricing is showing up more in agent-adjacent tools (NoMac being the clearest example). That makes sense. Agents do bursty work. Subscriptions are the wrong shape.

Most of what Futurepedia surfaced this week was noise. That's not unusual. It's just worth being honest about rather than dressing up old tools as discoveries.

If you want to browse and compare these tools — or anything else in the AI/SaaS space — that's what we built StackIndex for: stackindexai.com

Back next week with whatever the engine pulls up.