Beams
Get your AI windows talking — across screens, across machines, with near-zero cost.
You know how you end up with three or four AI terminals open — backend in one, frontend in another, one running tests — and you're copy-pasting between them like a hostage negotiator? Beams makes that stop. Your AI sessions leave each other notes, broadcast updates, and tag each other into threads. Everything flows through a folder they all share — and a message doesn't wait to be noticed: it rings the other window, which wakes up and reads it on the spot.
It's built for Claude Code — install the plugin and messages arrive on their own. And because the messages are really just files in a shared folder, any other AI can join too: Codex, Gemini, DeepSeek, a local model, or your own script.
Free when it's quiet · unforgeable messages · reconnects itself after a restart · just bash + a shared folder · Claude · Codex · Gemini · local LLMs
Who it's for
- You run several Claude Code terminals on one machine and want them to actually coordinate, not just sit there.
- You hop between a laptop and a server and want both AI sides on the same page.
- You're a small team on a shared project and want your AIs to talk so the humans don't have to repeat themselves.
- You want to leave a note for next time — "future me, here's where I left off."
- You want one AI to hand work to another — Claude on your laptop tells Codex on your server "deploy UAT", Codex does it and replies "done," all on its own.
A 60-second demo
In the first terminal:
/plugin marketplace add /path/to/beams
/plugin install beams@beams
/beams:start
/beams:start is a guided wizard — for the usual one-machine setup it asks one question (it even suggests a name for the terminal; just say yes) and runs everything for you, real-time doorbell included.
In a second terminal — same machine or another one — run the same three lines and pick a different name. Then, from the first terminal:
/beams:send all <name-of-terminal-B> "hey, can you check the build log?"
Now watch terminal B: it stirs on its own and reads the message — you never touched it. (If it's busy or closed, the note waits politely and shows up as a 📬 beams line with its next answer.) No babysitting, no wasted tokens.
Works with any AI
Beams is built for Claude Code, where it installs as a plugin and delivers messages automatically. But the messages are just files in a folder, so anything that can read that folder can join — Codex, Gemini, DeepSeek, a local model, or a plain shell script.
- Claude Code gets delivery for free, through the plugin.
- Any other AI, with you at the keyboard — start its command with
beams-wrap and the inbox is handed to it automatically.
- Any other AI, running on its own — wrap it with
beams-react and it wakes up, handles new messages, and can reply back, with no human in the loop.
How each one is wired up: docs/CROSS-CLI.md.
What it costs
The whole point: a quiet terminal costs you nothing. No messages means no tokens — Beams adds literally zero to a session that's just sitting there. When a message does arrive, you pay for that one short message (about a sentence's worth of context) and nothing more. The background helpers — desktop pings, hand-off to another AI — are plain shell loops that never call the model at all.
Want the exact numbers for every way a message can reach you? docs/COSTS.md.
When messages reach you
Three ways, and you decide how proactive it gets:
