New: Agent Harness

Bring your own AI agent framework to WhatsApp and every channel.

Choose your harness: MonoChat’s built-in agent, Claude Agent SDK, OpenAI Agents SDK, Pi, hosted Claude or OpenAI agents, or your own. Drop an AI Agent node into a flow and it answers on WhatsApp, Instagram, web chat and more, then hands over to your team with the full history.

Start free Talk to us
  • Claude Agent SDK
  • OpenAI Agents SDK
  • Pi
  • Custom endpoint

AI Harness

Choose a provider

  • MonoChat
  • OpenAI Agents SDK
  • Claude Agent SDK Selected
  • Pi Pi
  • OpenAI Agents API
  • Claude Managed Agents
  • </> Custom
AI Agent node Main · Fallback · Fast

How it works

Add a harness once, use it in any flow. The agent works the conversation and your team takes over whenever it should.

  1. 01

    AI Harness

    Add a harness once and choose the provider: built-in, an SDK, a hosted agent or your own endpoint.

  2. 02

    AI Agent node

    Drop the node into a flow, pick the harness, the main, fallback and fast models, and a budget per run.

  3. 03

    Every channel

    The flow runs on WhatsApp, Instagram, Messenger, TikTok, Telegram, web chat, SMS and voice.

  4. 04

    Human handover

    When a person is needed, the conversation moves to your shared team inbox with its full history.

The same harness also powers the agent-run API, so your own apps can start an agent run too.

Choose your harness

Seven harness providers, one inbox

Run the agent inside MonoChat, on MonoChat’s agent runner, in the provider’s cloud, or on your own server. Every option uses the same AI Agent node, the same channels and the same handover.

  • MonoChat

    Built-in harness

    Runs inside MonoChat

    Where it runs
    Inside MonoChat
    What you need
    Nothing; uses your MonoChat AI models

  • OpenAI Agents SDK

    OpenAI’s agent framework

    Runs inside MonoChat

    Where it runs
    Inside MonoChat
    What you need
    Nothing extra; uses your AI models

  • Claude Agent SDK

    The harness behind Claude Code

    Runs on MonoChat’s agent runner

    Where it runs
    MonoChat’s agent runner; each run is an isolated Claude Code-style agent process
    What you need
    Nothing extra; uses your AI models through MonoChat’s gateway
  • Pi

    Pi

    The Pi coding agent harness

    Runs on MonoChat’s agent runner

    Where it runs
    MonoChat’s agent runner
    What you need
    Nothing extra; models through MonoChat’s gateway

  • OpenAI Agents API

    Hosted by OpenAI

    Hosted by the provider

    Where it runs
    OpenAI cloud
    What you need
    Your OpenAI API key

  • Claude Managed Agents

    Hosted by Anthropic

    Hosted by the provider

    Where it runs
    Anthropic cloud
    What you need
    Your Anthropic API key
  • </>

    Custom

    LangGraph, Google ADK, CrewAI or your own code

    Your own

    Where it runs
    Your own server, any framework
    What you need
    An endpoint URL and a shared secret

Product names and logos belong to their owners.

Model roles

Three models per agent, and a budget for every run

Each AI Agent node has its own model roles, so a busy queue or a provider hiccup does not leave a customer waiting.

  • Main model

    The model that reasons, decides and writes the replies. Pick the one that fits the job best.

  • Fallback model

    Takes over after a soft limit or when the main model fails, so the conversation keeps going.

  • Fast model

    Handles background work, so the main model can focus on the conversation.

  • Budget limits

    Set a budget per run, so an agent never spends more than you planned on one conversation.

What you can build

Agents that work a case from first message to resolution, using your tools and your rules.

  • Customer support

    Look up orders, check delivery status, explain policies from your knowledge base and open a ticket or hand over when a case needs a person.

  • Sales and lead qualification

    Answer product questions, recommend options, collect the details your sales team needs and pass qualified leads to the right person.

  • Bookings and appointments

    Check availability through your own systems, offer slots, confirm or change bookings and send reminders through the flow.

  • Follow-ups

    Pick up where the conversation left off: chase a missing document, confirm a delivery or ask how a resolved case went.

Agent harness vs plain chatbot

A chatbot answers a message. An agent harness works the case.

Agent harness vs plain chatbot
Agent harness in MonoChatPlain chatbot
Steps per request As many as the task needs, within your budget One reply per message
Tools and systems AI function tools, MCP, knowledge base and custom functions Usually fixed menus or a single lookup
Framework Your choice: built-in, Claude Agent SDK, OpenAI Agents SDK, Pi, hosted or custom Fixed by the vendor
When a model fails Fallback model takes over Often no backup model
Channels Every channel MonoChat connects, from one flow Often one bot per channel
Handover Shared team inbox with full history Varies; context is often lost

Security and control

You decide what the agent may do

The agent runs inside your MonoChat flows, so the rules, tools and people around it stay yours.

  • Budgets per run

    Cap what a single run may spend, and let the fallback model take over after a soft limit.

  • Tools you choose

    Give the agent the MonoChat tools it needs: AI function tools, MCP servers, knowledge bases and custom functions.

  • Human handover

    Hand over to your team in the shared inbox with the full conversation history, so customers never repeat themselves.

  • Visible history

    Every agent reply stays in the conversation in the inbox, next to your team’s messages, for review and coaching.

  • Your keys, no AI markup

    Bring your own AI provider keys and pay the provider directly. MonoChat adds no AI markup.

  • Isolated runs

    On MonoChat’s agent runner, each Claude Agent SDK run is an isolated agent process.

FAQ

Agent Harness questions

Quick answers to the most common questions.

What is an AI agent harness?

An agent harness is the software layer around a language model that lets it act, not just reply: it runs the loop of reasoning, calling tools and checking results until the task is done. In MonoChat, an AI Harness is a service definition you add once; the AI Agent node in a flow, and the agent-run API, use it.

Which harness providers does MonoChat support?

MonoChat’s built-in harness, the OpenAI Agents SDK, the Claude Agent SDK, Pi, the OpenAI Agents API (hosted by OpenAI), Claude Managed Agents (hosted by Anthropic) and Custom, which connects your own server built with any framework, such as LangGraph, Google ADK, CrewAI or your own code.

Can I run a Claude Agent SDK or OpenAI Agents SDK agent on WhatsApp?

Yes. Add an AI Harness with the Claude Agent SDK or OpenAI Agents SDK as provider, add an AI Agent node to a flow and connect your WhatsApp number. The OpenAI Agents SDK runs inside MonoChat; the Claude Agent SDK runs on MonoChat’s agent runner, with each run as an isolated agent process. Both use your AI models, with no extra setup.

How do I connect my own agent built with LangGraph, ADK or CrewAI?

Choose the Custom provider and enter your endpoint URL and a shared secret. Your agent keeps running on your own server, and MonoChat brings it the conversations from every connected channel and handles handover to your team.

What are the main, fallback and fast models?

Each AI Agent node has three model roles. The main model reasons and replies; the fallback model takes over after a soft limit or a failure; the fast model handles background work. You can also set a budget limit per run.

Which channels does the agent work on?

Because the agent runs inside MonoChat flows, it works on every channel MonoChat connects: WhatsApp, Instagram, Messenger, TikTok, Telegram, web chat, SMS and voice.

Can the agent hand a conversation over to a person?

Yes. The agent can hand the conversation to your team in the shared inbox with the full history, so an agent picks up where the AI left off and the customer does not repeat themselves.

Do I need my own API keys, and does MonoChat add an AI markup?

The built-in harness, OpenAI Agents SDK, Claude Agent SDK and Pi use the AI models you already have in MonoChat. The hosted OpenAI Agents API and Claude Managed Agents need your own OpenAI or Anthropic API key. When you bring your own keys, MonoChat adds no AI markup.

Learn more about agent harnesses

First month of Growth on us

Choose your harness. Reach every channel.

Start free, connect WhatsApp and your other channels, and put your first AI Agent node to work today.

Start with $150 credit Chat on WhatsApp

One code per business • 30 days to redeem • No card needed

$150 to get started.

Claim