Commercial Agent Tools: October 24, 2026

1. Kavela Opens a Closed Beta for Cloud Agent Workflows With a Reviewer on Every Run

Kavela. Kavela opened its closed beta on October 23, pitching agent automations that “do not fail silently.” Users write a workflow in plain language, Kavela draws the steps, and each run gets an isolated cloud computer with persistent file context plus a reviewer that checks the output against a user-defined checklist. Users can connect an existing Claude or Codex subscription, and a Kavela MCP server promotes local sessions into repeatable cloud automations. Pricing is usage-based on agent minutes, starting at $30 a month (billed annually) for 3,000 workflow minutes, with a 14-day free trial and no card required. Source

2. Turm.dev Sells Pools of Durable Agent VMs on German Bare Metal, Managed Entirely Over SSH

Turm.dev. Turm.dev launched on Show HN on October 23 as “computers for your agents”: durable Linux VMs that keep their disks and run around the clock, created and managed through plain commands over SSH (ssh turm.dev new --name=...). Instead of buying one VPS per tool, customers buy a pool of CPU and RAM shared across all their VMs, which can come with Codex, Claude Code, OpenCode, Pi, Gemini or Aider, a full browser and a desktop preinstalled. VMs run on bare metal in Germany for EU data residency. Plans start at EUR 15 a month after a 7-day free trial (card required). Source

3. Corgi Multiplexes Claude Code, Codex, Pi and OpenCode Sessions in One Terminal

Corgi. Corgi, posted to Show HN on October 23 and inspired by Herdr, is a terminal multiplexer that shows what each coding agent is doing, which ones need input, and a running activity feed, with F1 jumping to the next unread agent. It offers a unified launcher for Claude Code, Codex, Pi and OpenCode across multiple accounts, session search and resume, per-workspace CPU, memory and GPU tracking, and token usage views, and it uses a cheap model to label links, worktrees and agent activity. It installs via a script on macOS, Linux and WSL; it is free to install, and its public repository does not publish an open-source license. Source

4. Sonah Reviews the Agent Session Behind a Change, Not Only the Diff

Sonah. Sonah, posted to Show HN on October 23 alongside its v0.3.0-beta.7 release, is a local code reviewer that reads a Claude Code or Codex session along with the code. It marks each ask in the prompt as done, partly done or not done, lists decisions the agent made on its own, flags claims contradicted by the session’s recorded results (such as “all tests pass” after a failing test), and runs a read-only code review agent, with every quote checked word for word. It runs through the user’s existing Claude or Codex login. Sonah is source-available under the Business Source License 1.1 and is free for personal projects and for organizations where up to three people use it. Source

5. Vericto Proxy Blocks Destructive SQL From Agents at the Database Wire Protocol

Vericto. Vericto posted its SQL proxy to Show HN on October 23 and shipped v4.16.0 on October 24. The proxy sits between clients (including AI agents) and a PostgreSQL or MySQL database on the native wire protocol, parses every statement into an AST, checks it against configured rules, and either forwards it or returns a normal SQL error, for example blocking a DELETE with no WHERE. Applications only change the host and port in their connection string. It is source-available under FSL-1.1-ALv2: free to self-host, including in commercial production, but not to resell as a competing service, with each release converting to Apache-2.0 after two years. Source

6. PlugVet Publishes Hygiene Grades for More Than 11,700 Agent Skills, Plugins and MCP Servers

PlugVet. PlugVet launched on Show HN on October 23 as an independent directory that grades Claude Code skills, Cursor and Codex plugins, and MCP servers from A to F, showing the checks behind each grade and the date it was assessed. It lists 11,739 graded entries so far, and F-graded entries are held back rather than published. The site says its grades are a hygiene check, not a security guarantee, and a scanner for grading any GitHub repo is listed as coming soon. The directory is free to browse and is funded through sponsor slots in a weekly newsletter. Source