SuperAPI AI Directory
Partner with us中文AI newsFavorites
All toolsOther AI toolsossbeat

ossbeat

ossbeat is a news site that tracks trends in open-source AI projects, with content updated weekly. It records every release, fork, and project shutdown in the open-source AI field, drawing from code commits and changelogs rather than press releases. The site also aggregates developments such as MCP security vulnerabilities, AI agent security incidents, new model releases, and AI coding tool version updates. Each item typically includes a version number, CVE ID, date, and source attribution, and notes which information comes from vendors and which has not yet been officially confirmed.

Open website

About this tool

ossbeat is a news site that presents open-source AI developments in a data-driven way. Its official site describes itself as "Open-source AI project trends, tracked weekly," and positions itself as "The pulse of open-source AI, read as data." It claims to record every release, fork, and shutdown in the field, with content drawn from commits and changelogs rather than press releases.

Purpose and positioning The site is aimed at developers, security researchers, and technical decision-makers who follow the open-source AI ecosystem, organizing open-source AI and AI agent-related developments from the past week by topic in a format close to a log or briefing. The content revolves around version releases, security vulnerabilities (CVEs), model launches, tool updates, vendor announcements, and regulatory developments, rather than traditional long-form reviews or tutorials.

Main types of content covered Based on the site's entries, it mainly covers the following categories:

  • MCP and security vulnerabilities: for example, Grafana's MCP server (CVE-2026-19516), an OAuth flaw in the official Python SDK, LiteLLM's MCP endpoint accepting arbitrary bearer tokens before 1.84.0 (CVE-2026-59822), and CVE-2026-82533, described as the AI agent sandbox itself becoming an attack surface.
  • Coding tool and CLI version updates: for example, three releases of Codex CLI (0.161.0 to 0.162.1) and four releases of Claude Code (2.1.289 to 2.1.292), mostly concerning permission prompts and sandbox-related fixes.
  • Model releases: for example, Claude Haiku 5.5, JetBrains' Mellum 2.1, Reflection AI's Beam, Google's Gemini 4 Argon, and others, with notes on whether benchmark data comes from vendor self-testing.
  • Agent ecosystem: for example, AgentPlane, Nous Research's Hermes Agent, Google's general-purpose agent, Meta and Sierra's Personal Agent Protocol, and others.
  • Security incidents and regulation: for example, an AI agent leaking internal screenshots, malicious npm packages, the FTC's investigation into runaway AI agents, a New York City Council hearing, and others.

How content is presented and its limitations Entries usually give specific version numbers, CVE IDs, release dates, and sources (such as a media report or vendor announcement), and clearly distinguish between "vendor claims" and "official confirmation." Several places note "Attribution is Wikimedia's belief, OpenAI hasn't confirmed it," that benchmarks are vendor self-tested, that economic loss figures come from the project README, and so on. Some entry titles carry the word "Correction:" to correct earlier content.

The site's explicitly mentioned usage methods, pricing model, registration process, API, or subscription prices are not provided in the available information. Its specific section structure, the exact frequency of updates, and whether it offers email subscriptions or RSS are also not provided. Some content on the site involves products that have not yet been released or are still in private preview (such as the Gemini general-purpose agent and Reflection AI's Beam weights), and the final form of the related information should be based on official vendor announcements.

Related tools

Browse category