Method InvokeAsync

Namespace

LMKit.Agents.Tools.BuiltIn.Net

Assembly

LM-Kit.NET.dll

InvokeAsync(string, CancellationToken)

Executes the tool with the specified JSON arguments.

public Task<string> InvokeAsync(string arguments, CancellationToken cancellationToken = default)

Parameters

arguments string

A JSON string representing an object that conforms to InputSchema. The agent runtime passes this directly from the model's tool call.

cancellationToken CancellationToken

A token to monitor for cancellation requests. Propagate this to any async operations within the tool.

Returns

Task<string>

A JSON string containing the tool's result. This is passed back to the model so it can incorporate the result into its response.

Examples

Example: Implementing InvokeAsync with proper error handling

public async Task<string> InvokeAsync(string arguments, CancellationToken cancellationToken = default)
{
    // Parse arguments
    SearchArgs args;
    try
    {
        args = JsonSerializer.Deserialize<SearchArgs>(arguments);
    }
    catch (JsonException ex)
    {
        throw new ArgumentException($"Invalid arguments: {ex.Message}", ex);
    }

    // Validate
    if (string.IsNullOrWhiteSpace(args.Query))
    {
        throw new ArgumentException("Query cannot be empty");
    }

    // Execute with cancellation support
    cancellationToken.ThrowIfCancellationRequested();
    var results = await _searchService.SearchAsync(args.Query, args.MaxResults, cancellationToken);

    // Return structured JSON
    return JsonSerializer.Serialize(new
    {
        query = args.Query,
        totalResults = results.Total,
        items = results.Items.Select(i => new { i.Title, i.Url, i.Snippet })
    });
}

Remarks

Implementation Guidelines:

• Parse arguments and validate against expected schema.
• Throw ArgumentException for invalid or malformed arguments.
• Throw HttpRequestException for network failures.
• Respect cancellationToken and throw OperationCanceledException when cancelled.
• Return compact, structured JSON. Avoid large blobs unless necessary.

Error Handling: Exceptions thrown from this method are caught by the runtime and converted to error results that the model can see. Design your error messages to be informative but safe for model consumption.