OpenClaw Releases OpenClaw 2.0: Guided Model Setup, 575 ms Control UI Startup, and One Trust Boundary Per Gateway
MarkTechPost Michal Sutter ● Covered by 6 sources
OpenClaw 2.0 rewrites setup and its browser control panel. The big catch: shared sessions are handy, but they’re not a security boundary.
Based on reporting by MarkTechPost, Michal Sutter — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
OpenClaw 2.0 is out, and the team clearly spent the break on plumbing instead of polish. After nearly seven weeks without a release, following 106 releases in the previous 230 days, this one rewrites installation, rebuilds the browser Control UI, and moves sessions and transcripts into SQLite.
The setup flow is the headline for anyone trying to get started without a scavenger hunt. OpenClaw can now look for AI access already on the machine, reuse verified Codex, ChatGPT, or Claude CLI sign-ins, accept an API key, launch a provider sign-in, or detect installed Ollama and LM Studio models. Before it saves anything, it checks that the chosen model can actually answer. Fresh OpenAI setups default to GPT-5.6.
On the local side, OpenClaw swapped out node-llama-cpp for a managed llama-server. Gemma 4 is now the RAM-gated llama.cpp default, and the llama.cpp default context was raised to 64K. The browser app is now the main surface, with conversations centered and files, approvals, and live work sitting alongside the chat instead of buried elsewhere.
The performance numbers are concrete. In a simulated default-chat test with a mocked Gateway and 50 ms HTTP/1.1 latency, JavaScript requests dropped from 140 to 45, and startup went from about 1.6 seconds to 575 ms. The new docked panels cover a workspace file editor, a git-backed Changes panel with pull-request status and CI summaries, a browser panel with element inspection and screenshot annotation, and a full-screen web terminal. But the guardrails stay tight: the file editor cannot create or delete files, Changes is read-only, and Create PR hands off to GitHub rather than filing inside OpenClaw.
There are also sharp edges around storage and rollback. Sessions and transcripts now live in SQLite, and the docs say downgrading is not free. Before going back to an older file-backed release, the current CLI has to restore archived legacy transcript artifacts, and sessions created after migration won’t show up in older versions. The release notes say to take a verified backup before upgrading.
Shared cloud sessions add real multiplayer, with a second person able to join live work or take it over with context intact. Owners and admins can choose whether that person may read, suggest changes, work in a draft, or participate directly. OpenClaw is blunt that this is not tenant isolation and not a security boundary, which is the sort of sentence more products should probably print in larger type.
My take — AI-written commentary, not fact-checked reporting
This is the right kind of AI release: less swagger, more controls and fewer surprises. OpenClaw is treating model choice, approvals, and sandboxing like actual product decisions instead of decorative safety stickers, which is refreshingly unglamorous. The shared-session feature is useful, but the explicit warning that it is not isolation is the part that shows some grown-up thinking.
Read more about this at: MarkTechPost