Configuration

As we have seen before, Eko accepts a parameter of type EkoConfig, which is defined as follows:

type EkoConfig = {
  llms: LLMs;
  agents?: Agent[];
  planLlms?: string[];
  callback?: StreamCallback & HumanCallback;
  defaultMcpClient?: IMcpClient;
  a2aClient?: IA2aClient;
};

This guide will walk you through configuring these parameters. You can also see EkoConfig’s Reference for code details.

EkoConfig.llms

EkoConfig.llms stores the large models available to Eko in a key-value format, where you can set the following for each model:

• Provider
• Model name
• API key
• Base URL
• TopK
• TopP

The models defined here will be used by the Planner and Agents. You can configure models with different performance levels to meet the varying needs of different components.

Here is an example that configures two models, with Claude 3.5 as the default and another one being GPT-4o:

let llms = {
  default: {
    provider: "anthropic",
    model: "claude-3-5-sonnet-20241022",
    apiKey: "sk-xxx",
    config: { topK: 5 },
  },
  openai: {
    provider: "openai",
    model: "gpt-4o",
    apiKey: "sk-xxx",
    config: { baseURL: "https://example.com/v1", topP: 0.7 },
  },
};

EkoConfig.agents

EkoConfig.agents describes the Agents available in the Eko workflow. Each Agent has its own name, description, toolkit, and available models.

For more information about Agents, please refer to the Agents section.

In this simple example, there is only one BrowserAgent:

import { BrowserAgent } from "@eko-ai/eko-extension";
let agents: Agent[] = [new BrowserAgent()];

EkoConfig.planLlms

EkoConfig.planLlms specifies the models that will be used by the Eko Planner. It is recommended to choose a high-performance model, as it will be responsible for planning the entire workflow. For example:

let llms = {
  default: {
    provider: "openai",
    model: "gpt-4o-mini",
    apiKey: "sk-xxx",
  },
  powerful: {
    provider: "anthropic",
    model: "claude-3-7-sonnet",
    apiKey: "sk-xxx",
  },
};
let eko = new Eko({ llms, planLlms: ["powerful"] });

EkoConfig.callback

EkoConfig.callback accepts a set of user-defined callback functions, which can include StreamCallback and HumanCallback.

• StreamCallback allows you to receive streaming updates on workflow progress, tool usage, and results.
• HumanCallback enables you to handle cases where human input or confirmation is required (e.g., when a user selects an option, confirms an action, or provides input).

For more information, please refer to the Callbacks section.

Example:

let callback = {
  onMessage: async (msg) => { /* handle streaming updates */ },
  onHumanInput: async (ctx, prompt) => { /* prompt user for input */ return "user input"; }
};

EkoConfig.defaultMcpClient

EkoConfig.defaultMcpClient is an instance of IMcpClient used to handle communication with an MCP (Multi-Component Platform) backend. Set this if your agents/tools require backend orchestration or state management.

EkoConfig.a2aClient

EkoConfig.a2aClient is an instance of IA2aClient for agent-to-agent communication. Use this if your workflow involves coordination or messaging between multiple agents.