OpenClaw unveils version 2.0 with enhanced model setup and security features

On August 31, 2026, OpenClaw launched its 2.0 version, introducing guided model setup, a revamped Control UI with a 575 ms startup time, and a single trust boundary per Gateway. The update follows a seven-week gap after 106 releases in the prior 230 days. Key improvements include rewriting installation, rebuilding the browser-based Control UI, and migrating sessions and transcripts to SQLite, with shared cloud sessions enabling real-time collaboration. The guided setup now identifies existing AI access on the machine, reusing verified Codex, ChatGPT, or Claude CLI sign-ins, or accepting API keys. It verifies the model’s capability before saving credentials, with fresh OpenAI setups defaulting to GPT-5.6. Local changes include replacing node-llama-cpp with a managed llama-server, setting Gemma 4 as the default for llama.cpp, and increasing the context size to 64K. OpenClaw emphasizes security, citing a 2026 crowdsourced attack test with 272K attacks across 41 scenarios, showing varying success rates against different models. The Gateway binds to loopback by default, and the platform warns that adaptive human attackers still exceed 80% success against current defenses, reinforcing the need for tool policies and sandboxing.