The blog is just launching — nothing published yet. Below are the first three posts we're writing on how the browser is built. Until they land, the code is the documentation: follow the repo.
Why we cloned the Anchor wire protocol byte-for-byte — anchor-api-key header, {data:{…}} envelopes, root-path CDP — so your SDK, Playwright, and Puppeteer code migrates with a base-URL swap.
Read →Sessions are claimed from a pre-warmed Object Pool, not cold-launched, and the CDP gateway splices WebSocket frames instead of parsing them. How both targets — ≤250ms warm create, ≤0.5ms proxy overhead — fall out of the design.
Read →One provider registry, four backends (Claude, Codex, Grok, Kimi) over API key, OAuth, or ACP — drive an already-logged-in CLI, no raw key. Plus the native MCP server driving Claude through a real lead-gen run.
Read →Dates are targets, not commitments — the drafts publish here when they're ready.
There's no separate changelog to drift out of date. Every shipped change — sessions, the CDP gateway, providers, the MCP server — lands as a tagged release on GitHub. That's the source of truth.
Open-source and free to self-host. Managed cloud is coming.