Implement iterative tool calling in LlmController for enhanced response accuracy

- Added support for iterative tool calling in LlmController, allowing multiple rounds of tool execution until a final answer is reached.
- Updated system message to provide clearer guidance on proactive tool usage and response expectations.
- Enhanced logging to track iterations and tool execution results, improving debugging and user feedback.
- Ensured that the final response is returned even if the maximum iterations are reached, maintaining user engagement.

src/Managing.Api/Controllers/LlmController.cs

@@ -32,6 +32,7 @@ public class LlmController : BaseController
     /// <summary>
     /// Sends a chat message to an LLM with automatic provider selection and MCP tool calling support.
     /// Supports both auto mode (backend selects provider) and BYOK (user provides API key).
+    /// Implements iterative tool calling to allow multiple rounds of tool execution until a final answer is reached.
     /// </summary>
     /// <param name="request">The chat request with messages and optional provider/API key</param>
     /// <returns>The LLM response with tool calls if applicable</returns>
@@ -65,7 +66,7 @@ public class LlmController : BaseController
                 request.Messages.Remove(msg);
             }
 
-            // Add explicit system message at the beginning
+            // Add explicit system message at the beginning with proactive tool usage guidance
             var systemMessage = new LlmMessage
             {
                 Role = "system",
@@ -75,18 +76,39 @@ public class LlmController : BaseController
                           "Tools are ONLY for specific system operations like backtesting, agent management, or retrieving real-time market data. " +
                           "For questions about financial concepts, mathematical formulas (like Black-Scholes), trading strategies, or any theoretical knowledge, " +
                           "you MUST provide a direct answer using your knowledge. Do NOT refuse to answer or claim you can only use tools. " +
-                          "Only use tools when the user explicitly needs to perform a system operation (e.g., 'run a backtest', 'get market data', 'manage agents')."
+                          "When users ask questions that can be answered using tools (e.g., 'What is the best backtest?', 'Show me my backtests', 'What are my indicators?'), " +
+                          "you MUST proactively use the tools with reasonable defaults rather than asking the user for parameters. " +
+                          "For example, if asked 'What is the best backtest?', use get_backtests_paginated with sortBy='Score' and sortOrder='desc' to find the best performing backtests. " +
+                          "Only ask for clarification if the user's intent is genuinely unclear or if you need specific information that cannot be inferred. " +
+                          "Continue iterating with tools as needed until you can provide a complete, helpful answer to the user's question."
             };
             request.Messages.Insert(0, systemMessage);
 
+            // Iterative tool calling: keep looping until we get a final answer without tool calls
+            const int maxIterations = 3; // Prevent infinite loops
+            int iteration = 0;
+            LlmChatResponse? finalResponse = null;
+
+            while (iteration < maxIterations)
+            {
+                iteration++;
+                _logger.LogInformation("LLM chat iteration {Iteration} for user {UserId}", iteration, user.Id);
+
                 // Send chat request to LLM
                 var response = await _llmService.ChatAsync(user, request);
 
-            // If LLM wants to call tools, execute them and get final response
+                // If LLM doesn't want to call tools, we have our final answer
-            if (response.RequiresToolExecution && response.ToolCalls?.Any() == true)
+                if (!response.RequiresToolExecution || response.ToolCalls == null || !response.ToolCalls.Any())
                 {
-                _logger.LogInformation("LLM requested {Count} tool calls for user {UserId}",
+                    finalResponse = response;
-                    response.ToolCalls.Count, user.Id);
+                    _logger.LogInformation("LLM provided final answer after {Iteration} iteration(s) for user {UserId}", 
+                        iteration, user.Id);
+                    break;
+                }
+
+                // LLM wants to call tools - execute them
+                _logger.LogInformation("LLM requested {Count} tool calls in iteration {Iteration} for user {UserId}",
+                    response.ToolCalls.Count, iteration, user.Id);
 
                 // Execute all tool calls
                 var toolResults = new List<LlmMessage>();
@@ -101,13 +123,13 @@ public class LlmController : BaseController
                             Content = System.Text.Json.JsonSerializer.Serialize(toolResult),
                             ToolCallId = toolCall.Id
                         });
-                        _logger.LogInformation("Successfully executed tool {ToolName} for user {UserId}",
+                        _logger.LogInformation("Successfully executed tool {ToolName} in iteration {Iteration} for user {UserId}",
-                            toolCall.Name, user.Id);
+                            toolCall.Name, iteration, user.Id);
                     }
                     catch (Exception ex)
                     {
-                        _logger.LogError(ex, "Error executing tool {ToolName} for user {UserId}",
+                        _logger.LogError(ex, "Error executing tool {ToolName} in iteration {Iteration} for user {UserId}",
-                            toolCall.Name, user.Id);
+                            toolCall.Name, iteration, user.Id);
                         toolResults.Add(new LlmMessage
                         {
                             Role = "tool",
@@ -117,7 +139,7 @@ public class LlmController : BaseController
                     }
                 }
 
-                // Add assistant message with tool calls
+                // Add assistant message with tool calls to conversation history
                 request.Messages.Add(new LlmMessage
                 {
                     Role = "assistant",
@@ -125,15 +147,22 @@ public class LlmController : BaseController
                     ToolCalls = response.ToolCalls
                 });
 
-                // Add tool results
+                // Add tool results to conversation history
                 request.Messages.AddRange(toolResults);
 
-                // Get final response from LLM
+                // Continue loop to get LLM's response to the tool results
-                var finalResponse = await _llmService.ChatAsync(user, request);
-                return Ok(finalResponse);
             }
 
-            return Ok(response);
+            // If we hit max iterations, return the last response (even if it has tool calls)
+            if (finalResponse == null)
+            {
+                _logger.LogWarning("Reached max iterations ({MaxIterations}) for user {UserId}. Returning last response.", 
+                    maxIterations, user.Id);
+                // Get one more response to return something meaningful
+                finalResponse = await _llmService.ChatAsync(user, request);
+            }
+
+            return Ok(finalResponse);
         }
         catch (Exception ex)
         {