Google open-sourced Agent Substrate, a Kubernetes runtime for scaling AI agent sandboxes on GKE with microVM or gVisor isolation, sub-500ms resume, and snapshotting of idle sessions. It targets coding-agent fleets and works with harnesses including Claude Code, Codex, Hermes, and Antigravity.
Published Oct 31 under MIT as a bundle for DeepSeek Harness, it watches one GitHub repo (and optionally Feishu), opens a work item when an allowed person asks, works in its own session and directory, and replies in the original thread, for example with a pull request, while a web panel shows what needs a human. DeepSeek labels it research code, not an official product, and setup is meant to be done by asking your coding agent to follow its SETUP.md…
As of Oct 31, Smithery's proxy, accounts, API keys, Connect, Toolbox, CLI and SDKs are gone, and smithery.ai stays only as a public directory now powered by Arcade.dev, with paid plans refunded back to Aug 5. Search your .mcp.json or .cursor/mcp.json for run.tools , server.smithery.ai and @smithery/cli and swap each for the vendor's official URL, like https://mcp.linear.app/mcp (migration list)…
Released Oct 31, pip install breakwise bundles a copy of the registry so breakwise explain BW 2026 0001 shows a notice offline: the package, the release that breaks your code, and a before and after fix that's run against real PyPI releases before it's accepted. A command that finds the code an upgrade breaks is next, and the notices are CC BY 4.0 if you want to feed them to your agent (GitHub)…
Released Oct 31, the agent's schedule tool now takes a tool call instead of a message, firing it directly at a set time, on idle or on a repeat ( every: "10m" ), with permissions and guardrails re checked at fire time and busy repeats skipping a tick instead of piling up. The release also tracks per session agent run history with main model costs and moves octohub and ChatGPT calls onto one persistent WebSocket, which is a breaking config change for per model transport (changelog)…
Launched Oct 30 on Show HN and Product Hunt, it lists each page's buttons, links and fields and has a small choice model (Jev) pick the next step, falling back to a cheap LLM when unsure, cites a page quote for every claim, and holds submit, pay, delete and send until your code allows them; try it with uvx fastbrowse "..." local or via its MCP server and agent skill. The makers report $0.030 per run versus $0.318 for Browser Use's hosted agent on their own 46 task suite, so treat that as self reported (Show HN recap)…
Spotted something we missed? Start a thread.
