Moonshot AI shipped Kimi WebBridge on May 14. A Chrome and Edge extension plus a local service lets any AI agent click, scroll, type, and screenshot pages — but your cookies and logged-in sessions never leave your laptop.
How the local pipe works
The extension talks to a daemon on your machine over Chrome DevTools Protocol, the same low-level interface developers use for debugging. Anthropic’s Claude for Chrome routes session content through their servers. WebBridge doesn’t. If you’re logged into Gmail or your bank, the agent sees what you see — locally.
Which agents plug in
Agent-agnostic by design. Out of the box: Kimi Code CLI, Claude Code, Cursor, Codex, Hermes. Anything speaking MCP or Computer Use can drive it. Typical use: scrape gated dashboards, fill forms behind login walls, run multi-step research across already-authenticated tabs without re-piping credentials anywhere.
Why it matters
Under the hood is Kimi K2.6 — 58.6% on SWE-Bench Pro, ahead of GPT-5.4 (57.7%) and Claude Opus 4.6 (53.4%). K2.6 fans out to 300 parallel sub-agents across 4,000 steps, so a single WebBridge task can split into hundreds of concurrent browser sessions. At $0.60 per million input tokens — roughly 12% of Claude Opus pricing — you’re getting an open-weight model beating it on coding. The privacy angle is real, but the cost gap is the bigger story.
You Might Also Like
- Kimi k2 6 Beats gpt 5 4 and Claude Opus 4 6 on swe Bench pro
- Openai Codex in Chrome Moves the Coding Agent Into Your Real Browser Session
- Anthropic Just Launched Code Review in Claude Code and 54 of prs now get Real Feedback
- Claude Replay Turns Your Anthropic Claude Code Sessions Into Shareable Video Like Replays
- Googles Chrome Devtools mcp Crosses 29k Stars 29 Tools That Give ai Agents Eyes in the Browser
Top AI Product 
Leave a comment