💬 Chat • 🧪 Workbench • 🏗️ AgentOS • 🗄️ DB • 🚦 CI/CD • 🔐 Self-Hosted • 🧑💻 Full code access
xpander.ai is the runtime and control plane to build, run, and ship reliable AI agents fast. Start in Chat to use your agents immediately. Deploy Managed Agents (serverless) or Embedded Agents (containers). Add tools in Workbench. Run in xpander cloud or your own VPC/Kubernetes. Every agent gets a stateful DB, CI/CD, and logs — with full code access. Works with any framework.
💬 Chat (chat.xpander.ai) — Generalist AI agent that auto‑discovers your agents and can schedule tasks. It’s the front door for everyone to use your agents and tools (supports your own domain and self‑hosting).
🧪 Workbench (app.xpander.ai) — Your control plane for any agent framework. Start with the Starter Kit template, then add tools from the menu: MCP servers, connectors, built‑in actions, or custom actions.
🏗️ AgentOS — Deploy agents of any framework on a production runtime in your VPC/Kubernetes or in xpander cloud; includes a reliable task scheduler and orchestration for long‑running jobs, plus stateful DB, logs, secrets, observability, and CI/CD.
🔌 Connector Hub — Generate and run MCP servers from OpenAPI specs or use the built‑in library of 2,000+ tools. Supports OAuth and API keys, and integrates natively with Claude, ChatGPT, and other MCP clients.
🛠️ AI Tools — OCR, browser automation, code interpreter, serverless code runner, PDF/CSV utilities, and more.
Sign up and log in at https://app.xpander.ai
Your account includes a preconfigured "Starter Kit" agent with a few tools and stateful storage to preserve sessions and showcase what’s possible. In Workbench, you can customize the system prompt, switch the model, or download the full agent code to modify and deploy.
Workbench screenshot
Then open Chat at https://chat.xpander.ai → pick the "Starter Kit" agent, add MCP tools, or ask it to run tasks.
This unified Chat is a rich generalist agent that can schedule tasks across all your deployed agents. Anything you add in Workbench shows up here automatically.
Example:
🛠️ Managed Agents (serverless, ~2 min)
- app.xpander.ai → New Agent → Starter Kit (or pick a framework)
- Add tools from the menu: MCP servers, connectors, built‑in actions, or custom actions
- Click Deploy → you get a stateful DB, logs, and CI/CD out of the box
workbench.mp4
🔧 Embedded Agents (container, ~3 min)
npm install -g xpander-cli && xpander login
mkdir hello-world && cd hello-world
xpander agent new --name "hello-world" --framework agno --folder .
python3 -m venv .venv && source .venv/bin/activate && pip install -r requirements.txt
xpander dev # local run
xpander deploy && xpander logsSwitch anytime
- Start Managed. Convert to Embedded by downloading code and redeploying:
xpander agent init <agent-id>
xpander agent deployWhere it runs
- xpander cloud or your VPC/Kubernetes (AgentOS) — for both Managed and Embedded
Managed vs Embedded (quick compare)
- Managed: serverless, no infrastructure, auto‑scales; fastest path for internal tools and quick production
- Embedded: your container and dependencies; full framework/runtime control; ideal for advanced policies, private packages, and GPUs
You can also build new MCP servers with xpander.ai — build once, use everywhere
- Generate from OpenAPI or bring your own
- Host in xpander cloud or your VPC/Kubernetes
- Use from agents and any MCP client (e.g., Claude Desktop)
🧩 Works with any framework: Agno, LangChain, PydanticAI, CrewAI — or your own runtime.
All agents—Managed or Embedded—are instantly available via Chat, Webhook, SDK, A2A, and MCP.
-
💬 Chat —
chat.<yourdomain>(or https://chat.xpander.ai) -
🌐 Webhook — per‑agent HTTPS endpoint
curl --location "https://webhook.xpander.ai?agent_id=$XPANDER_AGENT_ID" \
--header 'X-api-key: $XPANDER_API_KEY' \
--header 'Content-Type: application/json' \
--data '{"prompt":"Summarize Q3 pipeline risks"}'- 🧪 SDK
pip install "xpander-sdk[agno]"# .env
XPANDER_API_KEY=your_xpander_api_key
XPANDER_ORGANIZATION_ID=your_org_id
XPANDER_AGENT_ID=your_agent_idfrom xpander_sdk import Backend
from agno.agent import Agent
backend = Backend() # reads XPANDER_* from .env
agent = Agent(**backend.get_args()) # DB, MCP tools, system prompt
agent.print_response("What can you help me with?")-
🤝 A2A — agent‑to‑agent calls from Chat or the SDK. Schedule tasks across agents with one command.
-
🧩 MCP — governed connectors available to your agents and any MCP client (e.g., Claude Desktop). Host servers in xpander cloud or your VPC/Kubernetes.
Containerized agents include real code you control. A handler listens for tasks from the control plane and provides a Task object so you can choose the framework, tools, and runtime behavior. Not needed for serverless or no‑code agents.
Simply run xpander agent new and follow the wizard. It creates an xpander_handler.py file and a Dockerfile.
Here’s an example of xpander_handler.py already configured with Agno.
from xpander_sdk import Task, on_task, Backend
from agno.agent import Agent
@on_task
async def handler(task: Task):
backend = Backend(configuration=task.configuration)
agent = Agent(**backend.get_args(task=task))
return await agent.arun(message=task.to_message())- Local template:
02-agents/local-agent/— run the agent locally and use Ollama for a private LLM - DevOps agent:
02-agents/devops/— run the agent in the cloud with EKS permissions to troubleshoot logs - Travel agent:
02-agents/travel-agent/— a simple agent that helps with travel - Data agent:
02-agents/data-agent/— demo of building your own front‑end (instead of the xpander.ai Chat UI) with Streamlit
- Documentation: https://docs.xpander.ai
- API Reference: https://docs.xpander.ai/api-reference/07-sdk
- Join our Slack: https://xpander.ai/slack-community
- Open‑source SDK runtime: Apache 2.0
- Hosted platform: Commercial (free tier available)
