Yesterday's Top Launches: 5 Tools from October 9, 2026
Product Hunt’s October 9, 2026 lineup leaned toward workspace and developer tools that move AI assistants beyond chat and into existing files and systems, with IrisGo as Product of the Day, Figma’s canvas agent, and ButterShare’s direct WebRTC transfers among the launches.
Yesterday’s Product Hunt lineup was heavier on workspace tools than usual, with a couple of releases that try to answer the same question in different ways: what does an AI assistant look like once it stops being a chat window and starts living inside the files and systems you already use? Among the new developer tools and productivity apps that went live on October 9, 2026, five stood out for different reasons. IrisGo took the top spot on Product of the Day, Figma showed up with an agent that sits on the canvas, and a handful of smaller launches made convincing cases for their corners of the market.
ButterShare
Most file transfer tools make you upload first and ask questions later. ButterShare skips the middle entirely. It opens a direct WebRTC connection between two devices, streams the bytes over, and writes them straight to disk using the browser’s Origin Private File System. No account, no upload queue, no 2 GB ceiling that arrives right when you need to send a folder of 4K footage.
Chromium browsers and Firefox handle the disk-streaming path best; Safari works but falls back to memory, so very large transfers there will run into trouble. The interesting engineering detail is resumable transfers with checkpoints stored locally, which matters when a laptop sleeps halfway through a 20 GB send. It’s a small utility rather than a platform, but it solves a real annoyance cleanly. Worth bookmarking for anyone who regularly ships large assets to clients.
IrisGo for Solopreneurs
The premise here is blunt: if you run a business alone, the admin never stops landing on you, so hand it off. IrisGo is a desktop app for Windows and Mac that bundles a scheduled morning brief, an inbox triage layer called Radar, a plain to-do list, and a workflow recorder that watches you do something once and then replays it.
Radar is the part that earns its keep. It reads your inbox and surfaces only the bills, asks, and deadlines, each with a button to push the item onto your list. The demo day goes from 38 unread messages to three decisions in half an hour. Watch & Learn is the more ambitious bet, letting you demonstrate an expense report and then rerun it against a new receipt, though the accuracy of recorded workflows across messy real-world inputs is the thing to watch. It’s free in beta, backed by Andrew Ng’s AI Fund, and the makers are answering questions in the comments.
Databench by Alkera
Alkera is pitching something bigger than a notebook. Its open-source Databench workspace puts data engineering, analysis, and science into one multiplayer space where humans and agents edit the same .alknb.py files side by side. You can ask an agent to chart revenue by segment, watch it run cells, then have a teammate ask for a regional split and the agent edits the analysis in place.
Two features carry the trust argument. Column-level lineage traces a number from warehouse through transformation into the analysis, and knowledge entries show their sources and whether a human verified them. Sandbox environments let you rehearse pipeline changes before committing. The agent tooling is genuinely well thought out, and running cells on a separate GPU node means heavy training doesn’t lock up your session. The risk is scope: covering an entire data stack is a lot to promise, and the integrations list is long enough that depth in any one area is an open question.
Figma Agent
Figma’s entry works on the canvas itself rather than on exported images, which is the whole point. Because the agent reads your actual components and variables, generated layouts come back closer to production-ready instead of needing a cleanup pass. The bulk-editing angle is the most immediately useful: applying a design-system change across many screens in one go, or converting review comments directly into edits.
Prototyping from a prompt and adding motion to a design round out the feature set, and MCP integrations pull context from Notion, Slack, GitHub, and Linear so a design task can reference the issue or doc behind it. The skills feature, which saves repeatable workflows under a slash command, is the kind of thing teams either adopt fast or ignore entirely. No pricing details were shared, which is unusual for a launch of this size.
Rool
Rool takes the opposite approach to most AI products: instead of a chat window with your files bolted on, it gives you a full cloud machine where files, software, and memory persist between tasks. Everything is hosted in the EU, the models are self-hosted there too, and the company states flatly that it doesn’t sell your data or train on your content. Storage sits in Finland under EU law.
The persistent memory is the real differentiator. Because the machine keeps results and context, your next task starts from what you’ve already built rather than a blank conversation. Free gets you 240 AI credits a day and 1 GB per machine; Plus is €9 and Pro €25 with connectors and MCP on both. The privacy story is coherent and the pricing is honest, though anyone needing frontier-scale compute may find the credit ceilings tight.