Now under Vercel Labs with v1.0.2 (Oct 30), the Apache-2.0 Rust CLI (npm install -g agent-desktop) lets any coding agent snapshot a macOS, Windows or Linux app as JSON with a ref per element, act on a ref, and re-snapshot to verify, headless by default so it doesn't steal your pointer or focus. Snapshots can include private screen content, so keep them out of public logs (releases).
Since Oct 29, GET /v1/organizations/analytics/usage and /cost (beta header analytics usage cost 2026 08 24 , Admin or read:analytics key) break down tokens and USD by model, workspace, API key or user, though data lags 4–24 hours and the old Usage and Cost API is now legacy. Separately, since Oct 28 Messages and Token Counting requests can be 128 MB instead of 32 MB on the first party API only, while Bedrock, Vertex and Foundry keep lower caps (release notes)…
Released Oct 30, multi repository projects let one Claude Code, Codex or other agent conversation span several Git repos, submodules and local folders, with a per repo switcher for PRs and CI status (add them in Settings → Projects). It also adds one click scheduling of a continue message for after a Claude quota reset, and local MCP sign in that keeps credentials in your system keychain…
Released Oct 30, setting "experimental": { "artifacts": true } in opencode.json lets agents create, update and serve HTML or static artifacts locally or via plugin storage, while desktop terminals now belong to the session and keep their scrollback through server restarts. It also adds a guided Claude Agent provider setup, and same day v2.2.5 adds percentage compaction buffers like "20%" (changelog)…
Since Oct 30, the computer use tool is GA in both the Interactions and GenerateContent APIs: your loop sends a screenshot, gets back a UI action with an intent explaining it, executes it (e.g. with Playwright) and repeats. Responses can carry a safety decision of require confirmation or blocked, and Google still says to run it in a sandboxed VM or container (docs)…
Released this week under Apache 2.0, it sits between your agent and its model server on 127.0.0.1:8787 , blocking calls to undeclared tools or to hosts nobody mentioned in the conversation and running about 110 atif scan detectors for things like reading /proc/1/environ or curl noproxy , with Claude Code, OpenCode, Pi and three other harnesses tested in CI. Start with log only , and note it can't see URLs a script builds at runtime, so keep default deny egress in your sandbox (Tensor Journal)…
Spotted something we missed? Start a thread.