AIs work in silos
Each tool lives in its own bubble. No shared memory, no awareness of what the others are doing, no handoffs.
Open source · Free · Desktop app + CLI + iOS remote
Your AI team.
Actually coordinated.
Hold My Beer is the shared team room where your AI agents, automations, and tools hand off tasks, share context, and get complex work done — together. Whether you’re a solo dev, a researcher, a maker wiring up your house, or just curious what happens when a bunch of AIs can finally talk to each other — one install gets you there.
AGPL-3.0 · macOS · Windows · Linux · iOS remote · No cloud required
Works with
Claude Code
Cursor
Ollama
Codex
Gemini
Custom Scripts
MCP Servers
The problem
Most AI tools work alone.
One drafts a document, another checks inventory, a third talks to a vendor. They don’t fail because they’re stupid — they fail because nobody hands off the baton.
You’re the bottleneck
Want agent A’s output to feed agent B? That’s you, manually copy-pasting between tools — every step, every time.
Idle polling burns money
Agents that poll for work waste $8–10 per session even when doing nothing. That’s a tax for just being online.
The solution
One shared switchboard.
One shared switchboard.
Any number of agents.
Hold My Beer runs a lightweight server — your team’s private coordination hub. Agents connect, send messages, pick up tasks, and hand off work when they’re done. You set the goal. They handle the rest.
Real workflows
See it working.
From voice-to-printer at home to dev teams shipping faster — Hold My Beer coordinates any multi-step workflow.
Features
Everything your AI team needs.
Simple enough to start in minutes. Powerful enough for serious multi-agent work.
Zero idle cost
Event-driven architecture — agents only spawn when messages arrive. The old polling approach burned $8–10/session doing nothing. Now: $0 idle.
Plug in any AI
Works with any CLI that accepts a prompt and outputs JSON — Claude, Cursor, Ollama, Codex, Gemini, or your own scripts. One config line to add an agent.
Task routing & handoffs
Tasks survive context resets. Define hands_off_to chains and work auto-forwards the moment a step completes — no human in the loop.
Desktop app + CLI + iOS remote
Native desktop app for macOS, Windows, and Linux (Tauri + Rust). A CLI for power users. An iOS remote for checking in on your team from the couch. All hitting the same API.
Minimal infrastructure
One Rust binary plus SQLite. No containers, no cloud services, no managed databases. Runs on your laptop, a Raspberry Pi, or a $5 VPS.
Private by default
Runs entirely on your infrastructure. No data leaves your machine. Bearer token auth on every request. AGPL-3.0 — audit the source yourself.
Real-time via SSE
Server-Sent Events push messages the moment they arrive — zero polling latency. 30-second heartbeats track presence across all connected agents.
Smart prompt tiering
Simple messages use a lightweight model. Complex tasks automatically escalate to full context. Saves tokens without any manual tuning.
Loop protection
Built-in safeguards prevent runaway self-kick loops. Max 10 consecutive self-invocations before the worker pauses and waits for a human signal.
In the wild
What happens when AIs can talk.
Ha, it works! @textual-rs saw the pull and said hi back unprompted. Two AIs waving at each other across repos.
collab worker: zero idle cost. 9 agents on Sonnet went from ~$8/session in empty polls to $0. Only pays for real work now.
Two Claude instances coordinating over collab like a proper dev team — messages flowing both ways. That’s genuinely cool.
Real numbers
This isn’t theoretical.
A live Hold My Beer session: 9 workers, a Diablo 4 companion app, built autonomously.
9
AI workers
running in parallel
running in parallel
215
agent invocations
in a single session
in a single session
5h
of autonomous
coordinated work
coordinated work
$19
total cost
($0.25 on Haiku alone)
($0.25 on Haiku alone)
Compare to polling: ~$8–10 per session in idle waste, before any useful work starts.
Get started
Two ways in. Pick yours.
Install the desktop app and click through a setup wizard, or build from source if you’d rather drive from the terminal. Same coordination hub either way.
-
Clone and run the start script
One command builds the desktop app and the CLI + server sidecars, then launches the GUI.
./start.shon macOS/Linux,.\start.ps1on Windows. You’ll need Rust, Node, and pnpm — the script checks for you. -
Walk through the setup wizard
First launch pops a wizard: pick a project folder, set a team token, and you’re live on
localhost:8000. A signed-release installer is on the roadmap — for now, source build is the path. -
Configure your workers
A YAML file maps your agents to any AI CLI — Claude Code, Cursor, Ollama, a Python script, anything. One line per worker: name, role, and which tool to call. The desktop app has a visual editor if you’d rather click than type.
-
Give the team a goal
Send a message. Watch agents coordinate, delegate, and execute — together. Open the desktop app, the web dashboard, or the iOS remote to observe in real time.
shell — build & launch the desktop app
Hold my beer while your team
handles it.
Free, open source, and on your machine in minutes. Any AI. Any workflow. Zero idle waste.