Agents, including Codex sessions, can now search and read installed skills when a task needs one instead of loading the whole library into the prompt, and opt-in MCP Apps run in a pane beside chat with their own approval controls. Full backups can also go to external storage or Cloudflare R2 and restore into a fresh staging directory.
The MIT licensed, any model terminal coding agent now keeps one memory every project can read, syncs a navigation map at the end of each turn (view it with /map ), and shows elapsed time on MCP calls. It also makes Claude Haiku 5.5 its default Haiku tier and compacts tool results instead of truncating them…
Docker's YAML defined agent runtime ( docker agent run ) now ships built in toolsets for the current time, exact arithmetic, and random integer draws, and tells terminals whether an agent is idle, working, or waiting via OSC 7501. The release also fixes manifest budgets being ignored on API server runtimes and agent limits dropping when an A2A session resumes…
Since Oct 9, new Gemini Code Assist Standard and Enterprise plans can't be bought; existing ones keep working, can add licenses and renew annually until Jan 31, 2027, and monthly plans can renew until Dec 31, 2027. Google says agentic coding now lives in Antigravity, bundled with eligible Gemini Enterprise editions…
Every Veda app declares typed actions (70+ across the built in apps) that the agent runs through the same code path as your clicks, and the OS itself, not the model, holds sensitive or destructive actions until you approve them at the desktop. It's GPLv3, runs in QEMU, VirtualBox, or from a USB stick, and needs a Deepgram key for speech and the language model…
The TUI no longer updates itself: run /update to install a new version or move to OpenCode 2, or use opencode upgrade major from the shell. The release also allows OpenAI's ultrafast service tier and checks PTY connection origins and tickets before anything else…
Spotted something we missed? Start a thread.
