Run them anywhere. Leave them running.
Ground
ink paper
25,305 github stars 347,307 installs to date 508 community plugins 19 agent CLIs detected
the herd, from a client that isn't there click an agent
opencodeherdr+
❯ make the herdr.dev hero mock look exactly like real claude code
● Ha. I read the real pane over the socket — the mock was on v2.1.168, Claude Code is v2.1.198 now. Missing the logo block and the whole rate-limit row.
● Plan:
· index.html — swap ✻ for the real ▐▛███▜▌ mark, bump the version
· css/style.css — coral bypass line, add the 5h / Week bars
❯ perfect. and make the dialogue about this, obviously
⠏ Baking… (13m 49s · esc to interrupt)
❯
\~/Projects/herdr/website \> master \* ↑1 \> ctx ──────── 3% 31k/1M
⏵⏵ bypass permissions on (shift+tab to cycle)
\~/Projects/herdr master
❯ bun run dev
$ node scripts/prepare-docs.mjs && astro dev
02:10:44 \[types\] Generated 0ms
02:10:44 \[content\] Synced content
astro v5.18.1 ready in 668 ms
┃ Local http://localhost:4321/
Network use --host to expose
02:10:44 watching for file changes...
\+ Thought: 468ms
→ Read src/main.rs
\+ Thought: 292ms
This is src/main.rs — the entry point for herdr. Here's what it does:
1. Arg parsing — session::configure\_from\_args, then remote args for the SSH bridge.
2. Subcommand dispatch — server, client, update — each returns early.
3. Session mode — auto\_detect\_launch() connects to a running server or spawns one.
Build · MiMo V2.5 Free · 16.6s
Build · MiMo V2.5 Free OpenCode Zen
25.2K (13%) ctrl+p commands
claudedev server+
❯ finish the usage chart and ship it to staging
● Wrote src/components/UsageChart.tsx — stacked area, empty state, and the tooltip you asked for.
● Typecheck and build are clean, 41.2 kB gzipped. Ready to push to staging.
Bash command
bunx wrangler deploy --env staging
Deploy the dashboard to the staging environment
Do you want to proceed?
❯ 1. Yes
2. No, and tell Claude what to do differently (esc)
\~/Projects/web-dashboard feat/usage-charts
❯ bun run dev
VITE v6.3.1 ready in 412 ms
Local http://localhost:5173/
02:30:58 hmr update /src/components/UsageChart.tsx
02:31:04 hmr update /src/styles/chart.css
02:31:09 page reload src/routes/usage.tsx
02:31:15 WARN chunk "charts" is 512 kB after minification
02:31:18 hmr update /src/components/UsageChart.tsx
02:31:24 hmr update /src/components/Tooltip.tsx
02:31:33 ✓ built in 1.94s
codexbackfill+
• Explored
└ Read jobs/backfill\_events.py
• Ran
└ uv run pytest tests/test\_backfill.py
└ 14 passed · 0 failed · 41.3s
• Checkpointing works, so the job resumes from the last committed offset instead of restarting. The backfill in the next tab has been running for six hours and is unaffected by your laptop.
• Next I can add the p95 lag alert to the nightly run. Want that?
›
gpt-5.5 medium · \~/data-pipeline · backfill/events-v2 · Context 6% used
\~/data-pipeline backfill/events-v2
❯ uv run jobs/backfill\_events.py --from 2024-01-01
connected · batch size 5,000 · checkpoint every 60s
02:28:41 batch 4,102 ok 20,510,000 rows
02:29:03 batch 4,103 ok checkpoint written
02:29:25 batch 4,104 ok
02:30:07 batch 4,105 ok
02:30:18 WARN upstream throttled, retrying in 2s
02:30:42 batch 4,106 ok checkpoint written
02:31:09 elapsed 6h 12m · 68% complete · eta 2h 54m
The whole herd in one terminal. Click the sidebar, it's a real layout.
A real session: three workspaces, four agents, two of them mid-task. Nobody is attached to any of it right now — that's the point. Click an agent in the sidebar, it's a live layout, not an image.
01 Always running. → Herdr isn't an app you keep open. It's a server running in the background, and the terminals live inside it. Close the lid or drop the network and the agents keep working; restart the machine and Herdr brings the layout back and resumes their sessions. herdr-serverup 3 workspaces · 4 agents · nobody attached 02 You never hunt for the stuck one. → Herdr reads every pane and marks each agent working, blocked or idle. When one stops and needs an answer it says so, so you don't go pane by pane looking for whoever is waiting on you. web-dashboardblocked herdrworking data-pipelineidle 03 Agent-native. → The CLI and the socket API are the same surface agents drive. They split panes, start each other, prompt each other, and wait until another agent is genuinely blocked instead of firing keystrokes and hoping. $ 04 Runs what you already run. → Claude Code, Codex, Cursor, opencode, Grok and the rest. Herdr doesn't wrap them or replace them, it just owns their terminals — one binary, macOS and Linux, Windows in beta. Claude Code Codex Pi opencode Cursor Grok Copilot Hermes 19 agents detected out of the box
05
█████ ██████
████ █████████ █████████ ████ █ ████████ ████ ███ ██ ██ ██████ ███ ███ ██████ █████ █████ █████
Redacted. Something is coming.
$█████ █████ █████
███████ ███ █████ █████
Give them somewhere to live.
One command, and nothing else about your setup changes — the agent CLIs you already run keep working exactly as they did, except now they don't stop when you do.
`$curl -fsSL https://herdr.dev/install.sh | sh`copy
Windows beta: `irm https://herdr.dev/install.ps1 | iex` — all install methods →
pura: loading…
☰
📷 🔍 Discover ●▶ – ×
Ready
Close overlay
Hide for this session
Always hide on herdr.dev