diff --git a/engine/CMakeLists.txt b/engine/CMakeLists.txt
index c90be9f48..48f5a5ead 100644
--- a/engine/CMakeLists.txt
+++ b/engine/CMakeLists.txt
@@ -115,6 +115,7 @@ if(DEFINED CMAKE_JS_INC)
 
   add_library(${PROJECT_NAME} SHARED addon.cc
     ${CMAKE_CURRENT_SOURCE_DIR}/utils/cpuid/cpu_info.cc
+    ${CMAKE_CURRENT_SOURCE_DIR}/utils/file_logger.cc
     ${CMAKE_JS_SRC}
   )
 
@@ -131,6 +132,7 @@ if(DEFINED CMAKE_JS_INC)
 else() # Official build
   add_executable(${PROJECT_NAME} main.cc
     ${CMAKE_CURRENT_SOURCE_DIR}/utils/cpuid/cpu_info.cc
+    ${CMAKE_CURRENT_SOURCE_DIR}/utils/file_logger.cc
   )
 endif()
 
diff --git a/engine/main.cc b/engine/main.cc
index 1f403d869..1450d887c 100644
--- a/engine/main.cc
+++ b/engine/main.cc
@@ -6,6 +6,7 @@
 #include "utils/archive_utils.h"
 #include "utils/cortex_utils.h"
 #include "utils/dylib.h"
+#include "utils/file_logger.h"
 #include "utils/file_manager_utils.h"
 #include "utils/logging_utils.h"
 
@@ -29,16 +30,18 @@ void RunServer() {
   LOG_INFO << "Host: " << config.apiServerHost << " Port: " << config.apiServerPort << "\n";
 
   // Create logs/ folder and setup log to file
-  std::filesystem::create_directory(cortex_utils::logs_folder);
-  trantor::AsyncFileLogger asyncFileLogger;
-  asyncFileLogger.setFileName(cortex_utils::logs_base_name);
+  std::filesystem::create_directory(config.logFolderPath + "/" +
+                                    cortex_utils::logs_folder);
+  trantor::FileLogger asyncFileLogger;
+  asyncFileLogger.setFileName(config.logFolderPath + "/" +
+                              cortex_utils::logs_base_name);
+  asyncFileLogger.setMaxLines(config.maxLogLines);  // Keep last 100000 lines
   asyncFileLogger.startLogging();
   trantor::Logger::setOutputFunction(
       [&](const char* msg, const uint64_t len) {
-        asyncFileLogger.output(msg, len);
+        asyncFileLogger.output_(msg, len);
       },
       [&]() { asyncFileLogger.flush(); });
-  asyncFileLogger.setFileSizeLimit(cortex_utils::log_file_size_limit);
   // Number of cortex.cpp threads
   // if (argc > 1) {
   //   thread_num = std::atoi(argv[1]);
@@ -154,6 +157,18 @@ int main(int argc, char* argv[]) {
       RunServer();
       return 0;
     } else {
+      auto config = file_manager_utils::GetCortexConfig();
+      trantor::FileLogger asyncFileLogger;
+      asyncFileLogger.setFileName(config.logFolderPath + "/" +
+                                  cortex_utils::logs_cli_base_name);
+      asyncFileLogger.setMaxLines(
+          config.maxLogLines);  // Keep last 100000 lines
+      asyncFileLogger.startLogging();
+      trantor::Logger::setOutputFunction(
+          [&](const char* msg, const uint64_t len) {
+            asyncFileLogger.output_(msg, len);
+          },
+          [&]() { asyncFileLogger.flush(); });
       CommandLineParser clp;
       clp.SetupCommand(argc, argv);
       return 0;
diff --git a/engine/utils/config_yaml_utils.h b/engine/utils/config_yaml_utils.h
index 99db396cb..8e3668292 100644
--- a/engine/utils/config_yaml_utils.h
+++ b/engine/utils/config_yaml_utils.h
@@ -8,7 +8,9 @@
 
 namespace config_yaml_utils {
 struct CortexConfig {
+  std::string logFolderPath;
   std::string dataFolderPath;
+  int maxLogLines;
   std::string apiServerHost;
   std::string apiServerPort;
 };
@@ -16,6 +18,7 @@ struct CortexConfig {
 const std::string kCortexFolderName = "cortexcpp";
 const std::string kDefaultHost{"127.0.0.1"};
 const std::string kDefaultPort{"3928"};
+const int kDefaultMaxLines{100000};
 
 inline void DumpYamlConfig(const CortexConfig& config,
                            const std::string& path) {
@@ -27,7 +30,9 @@ inline void DumpYamlConfig(const CortexConfig& config,
       throw std::runtime_error("Failed to open output file.");
     }
     YAML::Node node;
+    node["logFolderPath"] = config.logFolderPath;
     node["dataFolderPath"] = config.dataFolderPath;
+    node["maxLogLines"] = config.maxLogLines;
     node["apiServerHost"] = config.apiServerHost;
     node["apiServerPort"] = config.apiServerPort;
 
@@ -48,8 +53,16 @@ inline CortexConfig FromYaml(const std::string& path,
 
   try {
     auto node = YAML::LoadFile(config_file_path.string());
+    int max_lines;
+    if (!node["maxLogLines"]) {
+      max_lines = kDefaultMaxLines;
+    } else {
+      max_lines = node["maxLogLines"].as<int>();
+    }
     CortexConfig config = {
+        .logFolderPath = node["logFolderPath"].as<std::string>(),
         .dataFolderPath = node["dataFolderPath"].as<std::string>(),
+        .maxLogLines = max_lines,
         .apiServerHost = node["apiServerHost"].as<std::string>(),
         .apiServerPort = node["apiServerPort"].as<std::string>(),
     };
diff --git a/engine/utils/cortex_utils.h b/engine/utils/cortex_utils.h
index 32dea9321..9673f0c1a 100644
--- a/engine/utils/cortex_utils.h
+++ b/engine/utils/cortex_utils.h
@@ -34,8 +34,8 @@ constexpr static auto kTensorrtLlmPath = "/engines/cortex.tensorrt-llm";
 
 inline std::string models_folder = "./models";
 inline std::string logs_folder = "./logs";
-inline std::string logs_base_name = "./logs/cortex";
-inline size_t log_file_size_limit = 20000000;  // ~20 mb
+inline std::string logs_base_name = "./logs/cortex.log";
+inline std::string logs_cli_base_name = "./logs/cortex-cli.log";
 
 inline std::string extractBase64(const std::string& input) {
   std::regex pattern("base64,(.*)");
diff --git a/engine/utils/file_logger.cc b/engine/utils/file_logger.cc
new file mode 100644
index 000000000..f684d457f
--- /dev/null
+++ b/engine/utils/file_logger.cc
@@ -0,0 +1,175 @@
+#include "file_logger.h"
+#include <algorithm>
+#include <iostream>
+#include <sstream>
+
+#ifdef _WIN32
+#include <io.h>
+#define ftruncate _chsize
+#else
+#include <unistd.h>
+#endif
+#include <string.h>
+
+using namespace trantor;
+
+FileLogger::FileLogger() : AsyncFileLogger() {}
+
+FileLogger::~FileLogger() = default;
+
+void FileLogger::output_(const char* msg, const uint64_t len) {
+  if (!circular_log_file_ptr_) {
+    circular_log_file_ptr_ =
+        std::make_unique<CircularLogFile>(fileBaseName_, max_lines_);
+  }
+  circular_log_file_ptr_->writeLog(msg, len);
+}
+
+FileLogger::CircularLogFile::CircularLogFile(const std::string& fileName,
+                                             uint64_t maxLines)
+    : max_lines_(maxLines), file_name_(fileName) {
+  std::lock_guard<std::mutex> lock(mutex_);
+  OpenFile();
+  LoadExistingLines();
+  TruncateFileIfNeeded();
+}
+
+FileLogger::CircularLogFile::~CircularLogFile() {
+  std::lock_guard<std::mutex> lock(mutex_);
+  CloseFile();
+}
+void FileLogger::CircularLogFile::writeLog(const char* logLine,
+                                           const uint64_t len) {
+  std::lock_guard<std::mutex> lock(mutex_);
+  if (!fp_)
+    return;
+
+  std::string logString(logLine, len);
+  std::istringstream iss(logString);
+  std::string line;
+  while (std::getline(iss, line)) {
+    if (lineBuffer_.size() >= max_lines_) {
+      lineBuffer_.pop_front();
+    }
+    lineBuffer_.push_back(line);
+    AppendToFile(line + "\n");
+    ++linesWrittenSinceLastTruncate_;
+    if (linesWrittenSinceLastTruncate_.load() >= TRUNCATE_CHECK_INTERVAL) {
+
+      TruncateFileIfNeeded();
+    }
+  }
+}
+void FileLogger::CircularLogFile::flush() {
+  std::lock_guard<std::mutex> lock(mutex_);
+  if (fp_) {
+    fflush(fp_);
+  }
+}
+
+void FileLogger::CircularLogFile::TruncateFileIfNeeded() {
+  // std::cout<<"Truncating file "<< totalLines_ <<std::endl;
+  if (!fp_ || lineBuffer_.size() < max_lines_)
+    return;
+
+  // Close the current file
+  fclose(fp_);
+  fp_ = nullptr;
+
+  // Open a temporary file for writing
+  std::string tempFileName = file_name_ + ".temp";
+  FILE* tempFile = fopen(tempFileName.c_str(), "w");
+  if (!tempFile) {
+
+    std::cout << "Error opening temporary file for truncation: "
+              << strerror(errno) << std::endl;
+    OpenFile();  // Reopen the original file
+    return;
+  }
+
+  // Write only the last max_lines_ lines to the temporary file
+  size_t startIndex =
+      lineBuffer_.size() > max_lines_ ? lineBuffer_.size() - max_lines_ : 0;
+
+  for (size_t i = startIndex; i < lineBuffer_.size(); ++i) {
+    fprintf(tempFile, "%s\n", lineBuffer_[i].c_str());
+  }
+
+  fclose(tempFile);
+
+  // Replace the original file with the temporary file
+  if (std::rename(tempFileName.c_str(), file_name_.c_str()) != 0) {
+    std::cout << "Error replacing original file with truncated file: "
+              << strerror(errno) << std::endl;
+    std::remove(tempFileName.c_str());  // Clean up the temporary file
+  }
+  //    else {
+  //     totalLines_.store(lineBuffer_.size() > max_lines_ ? max_lines_
+  //                                                       : lineBuffer_.size());
+  //   }
+
+  // Reopen the file
+  OpenFile();
+  // LoadExistingLines();
+  linesWrittenSinceLastTruncate_.store(0);
+}
+
+void FileLogger::CircularLogFile::OpenFile() {
+#ifdef _WIN32
+  auto wFileName = utils::toNativePath(file_name_);
+  fp_ = _wfopen(wFileName.c_str(), L"r+");
+#else
+  fp_ = fopen(file_name_.c_str(), "r+");
+#endif
+
+  if (!fp_) {
+// If file doesn't exist, create it
+#ifdef _WIN32
+    fp_ = _wfopen(wFileName.c_str(), L"w+");
+#else
+    fp_ = fopen(file_name_.c_str(), "w+");
+#endif
+
+    if (!fp_) {
+      std::cerr << "Error opening file: " << strerror(errno) << std::endl;
+    }
+  }
+}
+void FileLogger::CircularLogFile::LoadExistingLines() {
+  if (!fp_)
+    return;
+
+  // Move to the beginning of the file
+  fseek(fp_, 0, SEEK_SET);
+
+  lineBuffer_.clear();
+
+  std::string line;
+  char buffer[4096];
+  while (fgets(buffer, sizeof(buffer), fp_) != nullptr) {
+    line = buffer;
+    if (!line.empty() && line.back() == '\n') {
+      line.pop_back();  // Remove trailing newline
+    }
+    if (lineBuffer_.size() >= max_lines_) {
+      lineBuffer_.pop_front();
+    }
+    lineBuffer_.push_back(line);
+  }
+
+  // Move back to the end of the file for appending
+  fseek(fp_, 0, SEEK_END);
+}
+void FileLogger::CircularLogFile::AppendToFile(const std::string& line) {
+  if (fp_) {
+    fwrite(line.c_str(), 1, line.length(), fp_);
+    fflush(fp_);
+  }
+}
+
+void FileLogger::CircularLogFile::CloseFile() {
+  if (fp_) {
+    fclose(fp_);
+    fp_ = nullptr;
+  }
+}
\ No newline at end of file
diff --git a/engine/utils/file_logger.h b/engine/utils/file_logger.h
new file mode 100644
index 000000000..58b719019
--- /dev/null
+++ b/engine/utils/file_logger.h
@@ -0,0 +1,72 @@
+#pragma once
+
+#include <trantor/utils/AsyncFileLogger.h>
+#include <trantor/utils/Utilities.h>
+#include <atomic>
+#include <cstddef>
+#include <cstdint>
+#include <deque>
+#include <mutex>
+
+#ifdef _WIN32
+#include <windows.h>
+#else
+#include <unistd.h>
+#endif
+
+namespace trantor {
+
+class TRANTOR_EXPORT FileLogger : public AsyncFileLogger {
+ public:
+  FileLogger();
+  ~FileLogger();
+
+  /**
+     * @brief Set the maximum number of lines to keep in the log file.
+     *
+     * @param maxLines
+     */
+  void setMaxLines(uint64_t maxLines) { max_lines_ = maxLines; }
+
+  /**
+     * @brief Set the log file name.
+     *
+     * @param fileName The full name of the log file.
+     */
+  void setFileName(const std::string& fileName) {
+    filePath_ = "./";
+    fileBaseName_ = fileName;
+    fileExtName_ = "";
+  }
+  void output_(const char* msg, const uint64_t len);
+
+ protected:
+  class CircularLogFile {
+   public:
+    CircularLogFile(const std::string& fileName, uint64_t maxLines);
+    ~CircularLogFile();
+
+    void writeLog(const char* logLine, const uint64_t len);
+    void flush();
+    uint64_t getLength() const { return lineBuffer_.size(); }
+
+   private:
+    FILE* fp_{nullptr};
+    uint64_t max_lines_;
+    std::string file_name_;
+    std::deque<std::string> lineBuffer_;
+    std::atomic<int> linesWrittenSinceLastTruncate_{0};
+    static const uint64_t TRUNCATE_CHECK_INTERVAL = 1000;
+    mutable std::mutex mutex_;
+
+    void LoadExistingLines();
+    void TruncateFileIfNeeded();
+    void AppendToFile(const std::string& line);
+    void OpenFile();
+    void CloseFile();
+  };
+  std::unique_ptr<CircularLogFile> circular_log_file_ptr_;
+  uint64_t max_lines_{100000};  // Default to 100000 lines
+};
+
+}  // namespace trantor
\ No newline at end of file
diff --git a/engine/utils/file_manager_utils.h b/engine/utils/file_manager_utils.h
index e0bce0ca6..b4d7ab07a 100644
--- a/engine/utils/file_manager_utils.h
+++ b/engine/utils/file_manager_utils.h
@@ -135,7 +135,9 @@ inline void CreateConfigFileIfNotExist() {
   CTL_INF("Default data folder path: " + defaultDataFolderPath.string());
 
   auto config = config_yaml_utils::CortexConfig{
+      .logFolderPath = defaultDataFolderPath.string(),
       .dataFolderPath = defaultDataFolderPath.string(),
+      .maxLogLines = config_yaml_utils::kDefaultMaxLines,
       .apiServerHost = config_yaml_utils::kDefaultHost,
       .apiServerPort = config_yaml_utils::kDefaultPort,
   };
@@ -169,6 +171,27 @@ inline std::filesystem::path GetCortexDataPath() {
   return data_folder_path;
 }
 
+inline std::filesystem::path GetCortexLogPath() {
+  // TODO: We will need to support user to move the data folder to other place.
+  // TODO: get the variant of cortex. As discussed, we will have: prod, beta, nightly
+  // currently we will store cortex data at ~/cortexcpp
+  auto config = GetCortexConfig();
+  std::filesystem::path log_folder_path;
+  if (!config.logFolderPath.empty()) {
+    log_folder_path = std::filesystem::path(config.logFolderPath);
+  } else {
+    auto home_path = GetHomeDirectoryPath();
+    log_folder_path = home_path / config_yaml_utils::kCortexFolderName;
+  }
+
+  if (!std::filesystem::exists(log_folder_path)) {
+    CTL_INF("Cortex log folder not found. Create one: " +
+            log_folder_path.string());
+    std::filesystem::create_directory(log_folder_path);
+  }
+  return log_folder_path;
+}
+
 inline std::filesystem::path GetModelsContainerPath() {
   auto cortex_path = GetCortexDataPath();
   auto models_container_path = cortex_path / "models";
diff --git a/engine/utils/logging_utils.h b/engine/utils/logging_utils.h
index 77311c13e..fcaa3f4bb 100644
--- a/engine/utils/logging_utils.h
+++ b/engine/utils/logging_utils.h
@@ -30,4 +30,3 @@ inline bool log_verbose = false;
   } else {                         \
     std::cout << msg << std::endl; \
   }
-