AutoClaw, Open Analytics, and KerasFormers ship today

Three launches from August 22 put AI agents in three different places: the desktop, the analytics stack, and the model library. The useful question is what you can try now, and what is still too early to trust.

Z.ai launched AutoClaw on Product Hunt as a desktop work agent for multi-step tasks across files, browsers, office documents, web apps, and messaging. The product page says GLM-5.3 is built in; Z.ai’s own site lists office automation, browser automation, web product building, investment research, and integrations with services including WhatsApp, Telegram, Discord, and Lark. 12

The practical shift is from asking an agent for an answer to giving it a goal and receiving files, progress updates, and next steps in the same conversation. Z.ai says AutoClaw supports Windows 10 and later and both Apple Silicon and Intel Macs. The caution is authority: a tool that can browse, edit local files, and post into chats needs a test account, limited folders, and an approval step before it touches production data. 2

Open Analytics also launched on Product Hunt today. OpenLabs describes it as an open-source, privacy-first alternative to Google Analytics: a roughly two-kilobyte tracker, cookieless measurement, funnels, Core Web Vitals, Stripe revenue attribution, a read API, a CLI, and an MCP server for AI agents. 34

The interesting part for developers is the boundary. The documentation says the hosted service uses OAuth for the CLI and AI agents over MCP, while the repository says the full stack can be self-hosted under AGPL-3.0. The project’s self-hosting guide calls for Linux, Docker, four DNS records, about four gigabytes of RAM, and twenty-five gigabytes of disk. That makes it more than an analytics dashboard, but it also makes deployment and license review part of the decision. 45

A third developer-facing launch is KerasFormers, a Keras 3 model collection that reached Product Hunt today. The project’s repository lists pretrained models across vision, language, multimodal, speech, and depth tasks, with code licensed under Apache 2.0. The same README says the project is under active development and is migrating its release weights from GitHub to Hugging Face, so it recommends installing from the main branch while that migration continues. 67

That is the useful takeaway. KerasFormers is a portability layer for teams that want one Keras 3 implementation across JAX, PyTorch, and TensorFlow, not a new foundation model. Try it when framework portability is the problem. Pin a commit and test the weight-loading path before you make it part of a reproducible build, because the project itself says the package and latest models can be out of sync during the migration. 7

Today’s pattern is clear: AutoClaw expands what an agent can do on a workstation, Open Analytics gives agents a governed read surface for product data, and KerasFormers makes model code more portable. The right next step is not to hand any of them the keys. Give each one a narrow task, a pinned environment, and a way to inspect what happened.