My daily setup: Claude + OpenClaw + Tandem

[hydro13]

Sharing my daily workflow as the creator of Tandem — figured this is a good way to kick off Show and Tell.

The setup

I run Tandem as my only browser. Three agents connected simultaneously:

• Robin (me) — default workspace, normal browsing
• Claude — connected via MCP through Claude Code. Has its own workspace with 6+ tabs. I use it for research, code review in GitHub, reading docs, and composing emails in Gmail
• Kees (OpenClaw on Ollama) — runs locally, has its own workspace for autonomous tasks like job search automation and price monitoring

All three share the same browser session — same cookies, same history, same tabs. Tab locks prevent conflicts when multiple agents need the same site.

What a typical session looks like

1. I open Tandem, my sidebar has Telegram, WhatsApp, Discord, Gmail, Calendar pinned
2. Claude reads my open tabs via tandem_awareness_digest to understand what I'm working on
3. I ask Claude to research something — it opens tabs in its workspace, reads pages, summarizes
4. Meanwhile Kees monitors a few sites in the background via tandem_watch_add
5. If Claude hits a CAPTCHA or needs a judgment call, I see it in the Wingman panel and handle it

The key insight: the AI doesn't get a separate sandbox browser. It joins my browser. That means it has my authenticated sessions, my bookmarks, my context. No re-logging in, no API wrappers.

What makes it work

• Accessibility tree as API — every website works, no site-specific wrappers needed
• 8-layer security — web content can't attack the agent layer (prompt injection defense, script analysis, outbound guard)
• Stealth — websites see normal Chrome on macOS, not an automation tool
• Tab locks — prevents two agents from fighting over the same tab

Would love to see how others use Tandem. What's your setup?