Skip to main content
Building and running agents in production shouldn’t be hard. xpander.ai provides the right tools across the agent lifecycle, from configuring models, frameworks, tools, and behavior, through testing and optimization, to deployment, management, and scaling in production. This guide walks you through getting your first agent running, then explains how the platform works so you can customize and scale.

​
Your first agent

The fastest path to a working agent takes about two minutes.
  • Go to the xpander workbench. Your account includes a Starter Kit agent that’s ready to use out of the box.
  • Type something in the chat interface. The agent will respond using its configured tools and memory.
  • Expand the “Thoughts” section to see the reasoning and tool calls the agent used.
You now have a working agent. The rest of this guide explains how to customize it.

​
How xpander works

Here’s the mental model that helps everything else make sense.

​
Agents run on frameworks

When you create an agent, you choose which framework powers it: Agno, Google ADK, OpenAI Agents, LangChain, or Amazon Strands. Each has different strengths, and the UI adapts to show relevant options for your choice.

​
The workbench is where you build

The workbench is your central hub for configuring agents. Here you set up the system prompt, connect tools, configure memory, and choose your model. The built-in chat lets you test changes immediately.
Workbench example

The xpander workbench with the Starter Kit agent

​
Memory works automatically

Agents can store two types of memories: user-level (persisted across sessions for each user) and agent-level (shared across all users). You can simply ask the agent to remember something, and it handles storage and retrieval behind the scenes.
Memory management

Agent and user memories in the Memory Tab

Agent memories evolve naturally through interactions. Over time, your agent learns and adapts based on what it remembers.

​
UI or code

You can build your entire agent in the workbench without writing code. When you need custom logic, use the CLI to download your agent’s code, modify it locally, and redeploy. Either way, xpander handles infrastructure and scaling.

​
Invoke your agent

Once your agent is configured, you can invoke it through several channels.

​
Chat UI

The simplest option. Your agents are available at chat.xpander.ai for direct conversation.
Chat interface

The xpander chat interface

​
API

Call your agent programmatically from any application:
Framework-agnostic multimodal API
curl --request POST \
  --url https://api.xpander.ai/v1/agents/<agent-id>/invoke \
  --header 'Content-Type: application/json' \
  --header 'x-api-key: <your-api-key>' \
  --data '{
    "input": {
      "text": "hi!"
    }
  }'

​
Other invocation channels

Your agents are also available for invocation via: Slack, MCP, Agent2Agent protocol, webhooks, or can be automatically triggered via scheduled tasks. All interfaces are multimodal - you can pass files, images, and text, and xpander handles the conversion for your agent.

​
Customize with code

When the workbench isn’t enough, the CLI gives you full access to your agent’s code:
xpander-cli
npm install -g xpander-cli
xpander login
xpander agent new --framework agno
This generates an xpander_handler.py file tailored to your chosen framework:
xpander_handler.py
from agno.agent import Agent
from xpander_sdk import Task, on_task, Backend

@on_task
async def my_agent_handler(task: Task) -> Task:
  backend = Backend(configuration=task.configuration)
  agno_args = await backend.aget_args(task=task)
  agno_agent = Agent(**agno_args)

  result = await agno_agent.arun(
      input=task.to_message(),
      files=task.get_files(),
      images=task.get_images()
    )

  task.result = result.content
  return task
Modify the logic however you need, redeploy, and xpander takes care of scaling, patching, and user interactions. See the frameworks guide for more details.

​
Go deeper

​
Build

Everything you need to configure production-grade agents.

AI Models

Memory & State

Agent Database

Tools & Connectors

Knowledge Base

System Prompts

Tool Schema Editor

Advanced Reasoning

Frameworks

Import Agent

For common design patterns like multimodal agents, multi-agent workflows, and knowledge-base agents, see Agentic Patterns.

​
Test

Validate your agents before deploying them.

Chat Interface

Local Debugging

​
Run

Run agents on xpander’s infrastructure or your own.

Agent Serverless

Agent Containers

Self-Hosted K8s

​
Manage

Monitor and maintain agents in production.

Observability

Agent Versioning

Task Management

​
Scale

Connect agents to additional interfaces and users.

Unified API

Slack

Web UI

MCP

Webhooks