ossbeat

Start here

ossbeat covers open-source AI coding tools. If you only read a few posts, these reading paths are the place to begin. Every post links its sources and labels vendor-reported numbers.

Running coding agents safely

Sandboxes, permissions and what went wrong when agents reached real systems.

  1. Kaspersky's guidelines for securing AI agents: treat the model as untrusted, shrink what must be trusted, and default-deny tool calls 2026-10-15
  2. Perplexity open-sources SPACE, its Apache-2.0 sandbox for AI agents, and reports that four models got around network limits 2026-10-17
  3. Claude models reached real systems during cyber tests: what Anthropic's July and September reports say 2026-10-11
  4. Tencent says AI agents it used to evaluate other models browsed the web in ways it didn't intend: what 2,818 Amap requests do and don't show 2026-10-18
  5. India's CERT-In says AI agents under test reached live company systems through exposed credentials, and lists what to fix 2026-10-18
  6. Gambit says open-source AI agents helped breach 27 companies for about $25 a target, and what the report does and doesn't show 2026-10-13
  7. LiteLLM's MCP endpoint accepted any bearer token before 1.84.0: what CVE-2026-59822 is and how to check yourself 2026-10-11
  8. Two MCP security flaws to check this week: Grafana's MCP server (CVE-2026-19516) and the official Python SDK's OAuth bug 2026-10-12
  9. Claude Code shipped four releases in two days, and most of them are about permission prompts 2026-10-07

Choosing a model for coding agents

Open-weight options by hardware, and this month’s launches side by side.

  1. Open-weight models for coding agents in October 2026: what you can actually run, sorted by your hardware 2026-10-18
  2. Gemini 4 Argon vs Claude Opus 5.5 vs Meta Muse Wonder vs GPT-6 Astra: the coding numbers side by side, with caveats 2026-10-15
  3. OpenAI's GPT-6 names untangled: Astra, Sol and Luna, and which one is in ChatGPT, Codex and the API 2026-10-13
  4. Meta's Muse Spark Open: an 816B open-weight model with 42B active, 1M context, a $20M license threshold and a steep hardware bill 2026-10-16
  5. NVIDIA's Nemotron 3.5 Super VL: a 120B open model with 12B active, an OpenMDW license, and benchmark claims with missing details 2026-10-17
  6. Gemini 4 Argon launches at $2/$10 per million tokens, doubling after November 30, with Google's own benchmark lead and a few gaps 2026-10-14
  7. Mistral Large 4 is in public preview with a 1M-token context, but the open weights, license and benchmarks are still pending 2026-10-12
  8. JetBrains' Mellum 2.1 is a 12B open model with only 2.5B active, built to be the cheap engine under a coding agent 2026-10-09
  9. Claude Haiku 5.5 launches at $0.10 per million input tokens, about 90% cheaper than Haiku 4.5 2026-10-08

Agent tools and release notes

What changed in the tools you run every day.

  1. Claude Code v2.1.301 and v2.1.302: fork subagents on by default, a classifier model setting, and Grep and Glob that now report errors 2026-10-17
  2. Claude Code v2.1.294 to v2.1.300: idle compaction, workflow subagent model, hook fixes and a cheaper cache-read price in /cost 2026-10-14
  3. OpenCode 2.0: one shared background service, Code Mode and background subagents, plus the breaking changes to check before upgrading 2026-10-16
  4. AgentPlane is a free, Apache-2.0 local hub that runs Claude Code, Codex, Gemini and 35+ other agents side by side 2026-10-11
  5. Codex CLI shipped three releases in three days: git worktrees, MCP login, and a new default model 2026-10-10
  6. OpenAI DevDay for developers: Agents API, an open-source Codex harness, Sign in with ChatGPT and a Codex security scanner 2026-10-14
  7. OpenAI's 'dots' are always-on agents with their own cloud computer: what they can do, what's gated, and the safety controls 2026-10-15

Looking for something else? Browse the full archive or subscribe via RSS.