OpenClaw Releases OpenClaw 2.0: Guided Model Setup, 575 ms Control UI Startup, and One Trust Boundary Per Gateway

- OpenClaw 2.0 just landed — 16,000+ PRs, 933 contributors, and roughly half the repo's entire merge history crammed into one release. After seven weeks of radio silence, they shipped a guided setup that reuses your existing ChatGPT, Claude, or Codex credentials, verifies the model actually works before saving it, and defaults fresh OpenAI setups to GPT-5.6. The rebuilt browser Control UI cut startup from 1.6 seconds to 575 ms and slashed JS requests from 140 to 45. Sessions moved to SQLite, shared cloud sessions added real multiplayer, and the docs are refreshingly blunt: those session controls are not a security boundary. One trust boundary per gateway, model choice as your first prompt-injection mitigation, and a crowdsourced arena showing Claude Opus 4.5 resisting 99.5% of attacks while Gemini 2.5 Pro folds 8.5% of the time. Deployable for single-team use, not for multi-tenant products. Take a verified backup before upgrading — downgrades aren't free. Aggressive passive income for the contributors; for everyone else, read the manifest before you install.