LizardReview: hosting for coding agents

Skydive gives you an AI agent with its own computer in the cloud, including a browser, a terminal and its own files. You describe the job and talk to the agent like a coworker, over Slack, email, iMessage, the web app or a one-line command-line install. Each agent has a name, a role and memory, and can only use the tools you approve. Passwords go through a proxy, so the agent never sees them directly. The idea is a teammate that keeps working after you close your laptop, on things like support queues, calendars, GitHub and Notion, with more than 600 integrations. You can switch AI models mid-conversation and the agent keeps its name, memory and access. Starter is $20 a month for up to five people, plus usage at model cost with no markup. Team is $200. It went to #1 on Product Hunt in late August 2026.

Gumloop is a no-code automation tool where each step in a workflow is an AI agent. The steps can scrape a site, research a topic, pull data out of messy text, summarize, and make simple decisions. Think Zapier or Make, rebuilt with AI doing the work. It has more than 115 ready-made blocks, MCP support, scheduled runs and team controls. You drag blocks together to build things that would otherwise need code: process a pile of documents, research across several sources, turn unstructured text into a table, then send the results where they need to go. You pay in credits, so heavy workflows can get expensive. It's also harder to learn than basic automation tools. Gumloop assumes you already know what you want to automate and makes the AI parts easy. It won't teach you automation from scratch.

ClawsGO Science is a research agent with its own cloud computer. Give it a task in plain English, such as "pull the 2020–2026 trials comparing two treatments, run a meta-analysis, and draft the results section," and it searches the literature, screens papers, extracts the data into CSV tables, runs the statistics in Python, draws the forest plot, and writes the draft with citations. LaTeX manuscripts compile straight to PDF. Every search, command, and file edit is logged and replayable, so you can check the work instead of trusting it. You can point it at your own files with @, switch the underlying model per task, or connect a lab GPU over SSH for heavy jobs. It's aimed at researchers, but also at consultants, policy analysts, and due-diligence teams. The cloud machine isn't billed separately; you pay for model use in credits, and plan prices sit inside the app.

Hob is a free desktop assistant that runs entirely on your own computer. Install it and it picks an open model that fits your machine, such as Google's Gemma or OpenAI's gpt-oss, then downloads and sets it up. After that you use it like ChatGPT, with one difference: your chats and files never leave the laptop. No account, no subscription, no telemetry, and it keeps working offline. It does real file work. Point it at a folder and it can read a sales spreadsheet and chart revenue by region, fill out a PDF form, or edit a document. It asks before every change and lets you undo one it got wrong. On Mac, Mail and Calendar are built in. It needs 8 GB of RAM at minimum, 16 GB for everyday work, and 10 to 25 GB of disk. It runs on Apple silicon Macs, Windows, and Linux, and it's MIT-licensed by Frigade.