Publish an agent

Besides helping you internally, you can provide your agents to your customers. This allows you to build AI products, i.e. a support assistant which looks up the orders of a customer, or a domain specific assistant which uses your data. Since your agent uses your operations as tools, you can build such a product on top of your existing API.

There are three ways to provide an agent to your customers:

Option	Description
Developer portal	Your customers chat with the agent at the developer portal, no code required
Consumer API	You integrate the agent into your own app through the Consumer API
Agent-Call action	You provide the agent as a regular operation of your API

In all cases the customer must be authenticated, since every conversation is stored for a specific user.

Make an agent public

To provide an agent through the developer portal or the Consumer API, enable the Public setting of the agent. Only public agents are visible to your customers, the built-in agents are not public.

Developer portal

The developer portal contains a chat for all public agents. After login, a customer can select an agent at the account area and start a conversation. This is the fastest way to provide an agent, since you don't need to build any frontend.

Consumer API

If you want to integrate the agent into your own app, you can use the following endpoints of the Consumer API:

Method	Path	Description
GET	/consumer/agent	Returns all public agents
GET	/consumer/agent/:agent_id	Returns the details of an agent
GET	/consumer/agent/:agent_id/message	Returns the messages of the user
POST	/consumer/agent/:agent_id/message	Sends a new message to the agent

To send a message, the client sends a POST request with an access token of the user:

POST /consumer/agent/1/message
Authorization: Bearer [token]
Content-Type: application/json

{
  "item": {
    "type": "text",
    "content": "how can I solve 8x + 7 = -23"
  }
}

The response contains the answer of the agent and the id of the conversation:

{
  "id": "[conversation id]",
  "item": {
    "type": "text",
    "content": "Subtract 7 from both sides, then divide by 8: x = -15 / 4"
  }
}

To continue the conversation, pass the id of the conversation as previousId with the next message. The agent then receives the previous messages as context:

{
  "previousId": "[conversation id]",
  "item": {
    "type": "text",
    "content": "can you explain the second step?"
  }
}

Agent-Call action

With the Agent-Call action you can provide an agent as a regular operation of your API, i.e. POST /assistant. In contrast to the Consumer API, this gives you full control over the endpoint, i.e. the path, the scopes and the documentation. If the agent has an outgoing schema, the operation can directly return the structured output, so that your clients receive JSON in a defined format.

Costs

Every message causes costs at your LLM provider. If you provide an agent to your customers, you can pass these costs on through the costs setting of the agent. Customers then pay points per token, which they can buy through a plan, s. Monetization.