Your laptop stays yours
Each agent works inside its own microVM, with no network interface. Its commands never run on your host, and it never reaches another agent.
Open source · Apache 2.0
Live on DevHuntAgentPlaza gives each agent its own sealed machine, checks every tool call against laws you write, and holds the spend to a budget. In the morning you read a signed record of what happened, and nothing you did not allow.
Built by JUMPHASH, a member of the NVIDIA Inception program.
Orchestrator #200 plans the work and delegates it to the fleet · anthropic/claude-opus-5-5 · Ctrl+Enter to send
Every agent's running work is cancelled and nothing new starts until you resume. Queued work, VMs and workspaces are kept.
Any model, any engine
Every major provider through one gateway on the hub, which adds the key from its vault, so no key enters a VM. Twelve coding engines run inside each agent's own VM, under the same laws as everything else.
Modelsthrough the hub's gateway, with the keys in its vault
Coding enginesinside each agent's own VM · on by default
The problem
On your laptop today
In AgentPlaza
Each agent works inside its own microVM, with no network interface. Its commands never run on your host, and it never reaches another agent.
Every tool call is checked against laws you can read and change while the plaza runs. A call is allowed, refused, or sent to you to decide.
Each project runs on a budget in dollars or tokens, paid from escrow when the work checks out. STOP ALL holds every agent mid thought.
Every major provider and 12 coding engines, through one gateway. Keys live in the hub's vault and never enter a VM.
Start on the laptop you have. Add servers and Raspberry Pis as nodes later, each with a single use join code.
How it works
The installer checks every file against the release's SHA256SUMS. Start locally and nothing leaves the machine.
The orchestrator turns your goal into a plan, the planner writes the change, and the board hands each task to an agent in its own VM.

Risky calls wait for you. The plaza's own version control records every change with its intent, its review and its signature, then lands it.

See it run
Every flow below is the real web console against sample data. Each plays to its end, then hands over to the next.
Approve a held deploy. A release agent asked to deploy. The call waits on you; one click lets it run.
Proof
Every number here comes from the public repository and its CI. Read the code, run the suites, or watch a run.
4,011
tests run on Linux x86_64 for every change, beside Windows, macOS and arm64
3
acceptance suites: redteam, jail and killresume, against the containment rules
500
recent tool calls replayed by Check whenever you change a law
0
provider keys inside any agent's VM: the gateway adds them at the hub
Install
Pick your system. The installer checks each file against the release's SHA256SUMS, then puts AgentPlaza in your menu and agentplaza on your PATH.
PS> irm https://agentplaza.io/install.ps1 | iex Copy failed. Select the command and copy it by hand.
Windows 11, or Windows 10 2004 and later, with virtualization on in firmware. x64 runs agents on the Windows Hypervisor Platform.
$ curl -fsSL https://agentplaza.io/install.sh | sh Copy failed. Select the command and copy it by hand.
Apple silicon runs agents through Hypervisor.framework. Intel Macs run the hub and the desktop app.
$ curl -fsSL https://agentplaza.io/install.sh | sh $ sudo usermod -aG kvm "$USER"
Copy failed. Select the command and copy it by hand.
apt, dnf or pacman where it can, the AppImage elsewhere. Agents run in KVM microVMs, so your user joins the kvm group.
$ curl -fsSL https://agentplaza.io/install.sh | sh $ systemctl --user enable --now agentplaza-hub agentplaza-node
Copy failed. Select the command and copy it by hand.
Pi 4 or Pi 5 with 64 bit Raspberry Pi OS and 4 GB or more. A Pi 5 with 8 GB runs about 10 agents.
All downloads and SHA256SUMSor build from source with cargo build --release
FAQ
Yes. The hub, the nodes, the desktop app and the command line are open source under Apache 2.0. There is no account and nothing to subscribe to. You pay only your own model providers.
No. Each agent runs in its own microVM with no network interface. Its commands run only inside that VM, and its only link is vsock to the hub. The hub's egress proxy refuses loopback and private addresses, and records every connection.
Anthropic, OpenAI, Gemini, xAI, DeepSeek, Mistral, Kimi, GLM, OpenRouter, and local models through Ollama or LM Studio. 12 coding engines run inside the agent's own VM; Claude Code, Codex and OpenCode are on by default.
No. Start on the laptop you have: the hub and one node run there and nothing leaves it. Add machines later as nodes. A Pi 5 with 8 GB runs about 10 agents.
Press STOP ALL. It holds every agent mid thought, cancels the running work and refuses model calls until you resume. Then restart, wipe or retire the agent, or roll its VM back to a checkpoint.
Only what your model providers bill. Each project and task runs within a budget in dollars or tokens, tasks are paid from escrow when the work is verified, and every model call is metered.
The MSI is not signed yet. In SmartScreen choose More info, then Run anyway. Every installer checks each file against the release's SHA256SUMS before it installs.
Yes. Zed, JetBrains IDEs and Neovim drive plaza agents with agentplaza acp, and approvals come back to the editor. With a repo set, the work returns as a pull request.
Free under Apache 2.0. No account, nothing to subscribe to, and nothing leaves your machine until you host a hub.
Install AgentPlaza free