Skip to content
Eko Docs

Tools Hook

What is Eko Tools Hook?

Eko is a flexible framework that supports calling tools through workflows to achieve data orchestration and processing. To enhance the flexibility and customizability of this process, Eko provides a hook mechanism. Hooks allow users to intercept and process inputs and outputs before and after tool execution, thus optimizing workflow execution.

Why Tools Hooks?

In practical applications, we may need to make specific modifications or validations to the inputs or outputs of the tools. For example:

  1. Input Validation: Before tool execution, we can check if input parameters meet the expected format or range.
  2. Result Processing: After tool execution, we can format or transform the results to fit subsequent processes.
  3. Logging: We can log the tool execution process to facilitate workflow tracking.
  4. Conditional Processing: Modify inputs or decide whether to execute a tool based on specific conditions.

By using hooks, you can implement these functionalities without modifying the tools themselves, thus improving code reusability and the maintainability of workflows.

How to Use Hooks?

Implement the beforeToolUse and afterToolUse hook functions. These functions will be called before and after tool execution.

beforeToolUse

beforeToolUse?: (
  tool: Tool<any, any>,  // The tool currently being executed
  context: ExecutionContext,  // Execution context
  input: any  // Input parameters
) => Promise<any>  // Returns the modified input parameters

afterToolUse

afterToolUse?: (
  tool: Tool<any, any>,  // The tool currently being executed
  context: ExecutionContext,  // Execution context
  result: any  // Execution result
) => Promise<any>  // Returns the modified result

Usage Examples

Basic Usage

await eko.execute(workflow, {
  hooks: {
    beforeToolUse: async (tool, context, input) => {
      console.log(`Executing tool ${tool.name}, Input:`, input);
      return input;
    },
    afterToolUse: async (tool, context, result) => {
      console.log(`Tool ${tool.name} execution completed, Result:`, result);
      return result;
    }
  }
});

Practical Application Scenarios

Scenario 1: Parameter Validation and Conversion

await eko.execute(workflow, {
  hooks: {
    beforeToolUse: async (tool, context, input) => {
      // Validate numerical parameters
      if (input.amount && typeof input.amount !== 'number') {
        input.amount = Number(input.amount);
      }


      // Ensure date format is correct
      if (input.date && !(input.date instanceof Date)) {
        input.date = new Date(input.date);
      }


      return input;
    }
  }
});

Scenario 2: Error Handling and Retries

const MAX_RETRIES = 3;


await eko.execute(workflow, {
  hooks: {
    beforeToolUse: async (tool, context, input) => {
      context.retryCount = 0;
      return input;
    },


    afterToolUse: async (tool, context, result) => {
      if (result.error && context.retryCount < MAX_RETRIES) {
        context.retryCount++;
        console.log(`Retrying ${context.retryCount} time(s)`);
        return await tool.execute(context, input);
      }
      return result;
    }
  }
});

Scenario 3: Performance Monitoring

await eko.execute(workflow, {
  hooks: {
    beforeToolUse: async (tool, context, input) => {
      context.startTime = Date.now();
      return input;
    },


    afterToolUse: async (tool, context, result) => {
      const duration = Date.now() - context.startTime;
      console.log(`Tool ${tool.name} execution time: ${duration}ms`);


      // Send performance data to monitoring system
      await metrics.record({
        tool: tool.name,
        duration,
        success: !result.error
      });


      return result;
    }
  }
});

Scenario 4: Data Encryption for Sensitive Information

await eko.execute(workflow, {
  hooks: {
    beforeToolUse: async (tool, context, input) => {
      // Encrypt sensitive data
      if (input.phoneNumber) {
        input.phoneNumber = encrypt(input.phoneNumber);
      }
      if (input.password) {
        input.password = encrypt(input.password);
      }
      return input;
    },


    afterToolUse: async (tool, context, result) => {
      // Clean up sensitive information from the returned result
      if (result.userInfo) {
        delete result.userInfo.token;
        delete result.userInfo.bankAccount;
      }
      return result;
    }
  }
});

Next Steps

Now that you understand the concept of tool hooks, let’s explore the more powerful hook system in Eko to learn how to implement human-in-the-loop with LLMs: