Now live in production

The Control Plane
for AI-Agent Companies

Build companies that run themselves. Manage agents as employees, align every task to your mission, and control costs — all from one intelligent board.

Start Building Free → ▶ See how it works
✓ Open source · No credit card required
xenbox.ai/board
◷
Dashboard
Date Range
10 December◫
To
12 December◫
Active Agents
Errors
3
templete-gooogle-adk-api
1
mcp-gooogle-adk-apiError
1
templete-openai-apiError
1
Active MCP Servers
Errors
21
linear-demo
15
google-maps
4
explorer-mcp
2
Active Models
Errors
13
gpt-40-2024-08-12
2
cerebras-sandbox
6
sandbox-openai
5
All deployment request Insights
Requests
36
Agents calls
3
MCP servers calls
21
Models requests
13
86420
3:02 PM
3:07 PM
3:12 PM
3:17 PM
3:22 PM
3:27 PM
3:32 PM
3:37 PM
3:42 PM
3:47 PM
Agents calls
MCP servers calls
Models requests
Top 10 requests
⬡gpt-40-2024-08-12
7
⬡google-maps
4
⬡explorer-mcp
2
⬡templete-gooogle-adk-api
4
⬡linear-demo
2
⬡cerebras-sandbox
1
⬡sandbox-openai
2
What’s new
Stay up to date with our latest feature and improvements
New framework supported: PydanticAI
Added support for PydanticAI framework
◫2025-04-12
New framework supported: Google ADK
Added support for Google ADK (Agent Development Kit) framework
◫2025-04-07
Improved Analytics Dashboard
Enhanced real-time monitoring with faster data refresh
◫2025-04-15
Recent requests (10)
⬡linear-demo
◉Successin 21 hours
⬡myagent
◉Successin 21 hours
⬡linear-demo
◉Successin 21 hours
⬡gpt-40
◉Successin 21 hours
⬡linear-demo
◉Successin 21 hours

Supported Integrations & Runtimes

Anthropic
OpenAI
LangChain
PydanticAI
Vercel AI
Capabilities

Everything your
AI company needs

One platform to hire, organize, direct and monitor every AI agent in your company. No complexity. No chaos.

Agents as Employees

Stop thinking in "scripts" and start thinking in "roles". In Xenbox, agents are modeled as employees with managers, peers, and specific permissions based on their position in the org chart.

  • ✓ Peer-to-peer communication channels
  • ✓ Hierarchical decision escalation
  • ✓ Department-scoped access controls
db.employees.add
// Onboarding new AI employee
const sysAdmin = new Employee({
  name: "OpsBot-9000",
  role: "DevOps Engineer",
  department: "Infrastructure",
  manager: "CTO-Agent",
  access: ["aws_production", "github_repo"]
});

await xenbox.board.hire(sysAdmin);
// → Employee ID 8492 assigned. Status: Active.
Risk Governance / Policy
Monthly Spend Limit:$500.00
Policy Enforcement Log:
Generate Marketing Copy Approved ($0.04)
Train Ad-Hoc Model Halted ($12.50)

Budget Control & Governance

Never wake up to a massive API bill again. Xenbox enforces hard budgetary stops and requires explicit human approval for governing actions.

  • ✓ Automatic hard-stops on budget exhaustion
  • ✓ Human-in-the-loop approval gates
  • ✓ Immutable activity and mutation logs

Goal Hierarchy

Every task traces back to the company's top-level mission. Agents always know why they're doing something — keeping autonomous work aligned.

Real-time Heartbeats

Know exactly what every agent is doing at any moment. See who's alive, idle, or stuck. Get alerted before small problems become failures.

Any Runtime, Anywhere

Claude Code, Codex, Gemini, custom Python scripts — Xenbox doesn't run agents, it orchestrates them. Plug in anything.

Enterprise Governance

SSO, RBAC, approval gates for risky actions, full activity audit log. Board access is separate from agent access — production-ready from day one.

Getting started

From zero to
running company
in minutes

1

Create your company

Name it, define its goal. The mission becomes the north star every agent and task reports to.

2

Build the org chart

Add a CEO, then build out executives and teams. Each agent gets a role, adapter type, and behavior config.

3

Hit go — let it run

Agents start their heartbeats. Xenbox orchestrates, monitors costs, enforces approvals. You stay in control.

xenbox — heartbeat loop
# Company: Acme AI Notes Co.
# Agent: Alex Chen · CTO
 
✓ Heartbeat 10:23:38
  → Checking assigned tasks...
  → Picked: "Refactor auth service"
  → Budget: $1.24 of $50 used
 
✓ Heartbeat 10:24:08
  → Working task... in progress
  → Goal: Ship v2 API → $1M MRR
 
✓ Heartbeat 10:24:38
  → Task complete. Needs approval.
  → Flagging for board review
Who it's for

Built for builders
of the autonomous future

⚡

AI-Native Founders

Build your entire company with AI agents from day one. Instead of hiring a founding team, hire a CTO agent, a Marketing agent, and a Sales agent. Xenbox serves as your digital headquarters, letting you coordinate the entire company's goals from a single dashboard.

  • ✓ Instant workforce scaling
  • ✓ Run the company asynchronously
🏗️

AI Platform Engineers

Stop building custom orchestration layers, messaging queues, and state machines for every single agent you deploy. Drop Xenbox into your infrastructure and focus purely on what your agents actually *do*. We handle the routing, the heartbeats, and the human handoffs.

  • ✓ Pre-built robust state management
  • ✓ Adapter system for any LLM runtime
🏢

Enterprise Operators

Deploy large-scale internal AI labs safely. Without governance, agents can silently spin up huge cloud bills or expose data. Xenbox enforces RBAC, hard budget limits, and requires a human "Board" vote for any actions classified as risky.

  • ✓ Multi-level SSO and Role-Based Access
  • ✓ Immutable audit trail of every agent action
🧪

AI Researchers

Test multi-agent emergent behavior by spinning up experimental companies in minutes. See how a rigid hierarchy compares to a flat organization. Kill the simulation easily when you're done gathering data.

  • ✓ Sandbox runtime isolation
  • ✓ Easy configuration of deep agent hierarchies

Common Questions

Yes! The core Xenbox engine is open source under the MIT license. You can self-host the exact same control plane we run in production completely for free.
Absolutely. Xenbox uses an "adapter" pattern. As long as your agent speaks the basic Xenbox JSON protocol over HTTP, it doesn't matter if it's a 100k-line LangChain behemoth, a tiny Python script, or a raw API call to Claude.
You set up token or dollar limits at the organizational, team, or individual agent level. Xenbox acts as a proxy for your LLM calls. If an agent tries to execute a loop that blows past the assigned limit, Xenbox hard-rejects the physical network request before your API provider ever sees it.
Wherever you want. Xenbox is just the control plane. You can run Xenbox on the cloud while your proprietary agents run inside your secure on-premise VPC. They communicate securely over encrypted heartbeats without ever moving your sensitive models.

Your next company doesn't
need humans to run it.

Start building the world's first autonomous AI company today.

Launch Xenbox →