diff --git a/DESCRIPTION b/DESCRIPTION
index 2930c75..9c62a60 100644
--- a/DESCRIPTION
+++ b/DESCRIPTION
@@ -1,6 +1,6 @@
 Package: finnts
 Title: Microsoft Finance Time Series Forecasting Framework
-Version: 0.6.0.9016
+Version: 0.6.0.9017
 Authors@R: 
     c(person(given = "Mike",
            family = "Tokic",
diff --git a/NEWS.md b/NEWS.md
index 87504ff..d2beb0a 100644
--- a/NEWS.md
+++ b/NEWS.md
@@ -1,4 +1,4 @@
-# finnts 0.6.0.9016 (development version)
+# finnts 0.6.0.9017 (development version)
 
 ## Improvements
 - TimeGPT Integration
diff --git a/R/agent_summarize_models.R b/R/agent_summarize_models.R
index 678f887..7abc185 100644
--- a/R/agent_summarize_models.R
+++ b/R/agent_summarize_models.R
@@ -119,7 +119,8 @@ summarize_models <- function(agent_info,
     "theta" = summarize_model_theta,
     "xgboost" = summarize_model_xgboost,
     "nnetar-xregs" = summarize_model_nnetar,
-    "prophet-xregs" = summarize_model_prophet
+    "prophet-xregs" = summarize_model_prophet,
+    "timegpt" = summarize_model_timegpt
   )
 
   # Setup parallel processing
@@ -4807,7 +4808,6 @@ summarize_model_snaive <- function(wf) {
 }
 
 
-
 #' Summarize an STLM-ARIMA Workflow
 #'
 #' Extracts and summarizes key information from a fitted STLM-ARIMA workflow.
@@ -7761,6 +7761,125 @@ summarize_model_xgboost <- function(wf) {
   )
 }
 
+#' Summarize a TimeGPT Workflow
+#'
+#' Extracts and summarizes key information from a fitted TimeGPT workflow,
+#' including model arguments, engine parameters, and api configuration details.
+#'
+#' @param wf A fitted tidymodels workflow containing a timegpt_model()
+#'   model with engine 'timegpt_model'.
+#'
+#' @return A tibble with columns: section, name, value containing model details.
+#'
+#' #' @noRd
+summarize_model_timegpt <- function(wf) {
+  digits <- 6
+  scan_depth <- 6
+
+  if (!inherits(wf, "workflow")) stop("summarize_model_timegpt() expects a tidymodels workflow.")
+  fit <- try(workflows::extract_fit_parsnip(wf), silent = TRUE)
+  if (inherits(fit, "try-error") || is.null(fit$fit)) stop("Workflow appears untrained. Fit it first.")
+
+  spec <- fit$spec
+  engine <- if (is.null(spec$engine)) "" else spec$engine
+  if (!identical(engine, "timegpt_model")) {
+    stop("summarize_model_timegpt() only supports timegpt_model() with engine 'timegpt_model'.")
+  }
+
+  .format_numeric <- function(x) {
+    if (is.character(x) && (x == "auto" || x == "")) {
+      return(x)
+    }
+    x_num <- suppressWarnings(as.numeric(x))
+    if (!is.na(x_num) && is.finite(x_num)) {
+      rounded <- round(x_num, digits)
+      if (rounded == floor(rounded)) {
+        return(as.character(as.integer(rounded)))
+      }
+      return(format(rounded, scientific = FALSE, trim = TRUE))
+    }
+    as.character(x)
+  }
+
+  # Predictors / outcomes
+  mold <- try(workflows::extract_mold(wf), silent = TRUE)
+  preds_tbl <- .extract_predictors(mold)
+  outs_tbl <- .extract_outcomes(mold)
+
+  # Flag external regressors (non-date predictors)
+  xreg_names <- character()
+  if (!inherits(mold, "try-error") && !is.null(mold$predictors) && ncol(mold$predictors) > 0) {
+    is_date <- vapply(mold$predictors, function(col) inherits(col, c("Date", "POSIXct", "POSIXt")), logical(1))
+    xreg_names <- names(mold$predictors)[!is_date]
+    if (length(xreg_names) > 0 && nrow(preds_tbl) > 0) {
+      preds_tbl$value[preds_tbl$name %in% xreg_names] <-
+        paste0(preds_tbl$value[preds_tbl$name %in% xreg_names], " [regressor]")
+    }
+  }
+
+  # Recipe steps
+  # no significant step
+  steps_tbl <- tibble::tibble(section = character(), name = character(), value = character())
+
+  arg_names <- c("finetune_steps", "finetune_depth")
+  arg_vals <- vapply(arg_names, function(nm) .format_numeric(.chr1(spec$args[[nm]])), FUN.VALUE = character(1))
+  args_tbl <- tibble::tibble(section = "model_arg", name = arg_names, value = arg_vals)
+
+  # Engine params
+  eng_tbl <- tibble::tibble(section = character(), name = character(), value = character())
+
+  # Locate the timegpt fit object
+  is_timegpt_fit <- function(o) {
+    inherits(o, "timegpt_model_fit") || (is.list(o) && !is.null(o$train_data) && !is.null(o$forecast_horizon))
+  }
+  timegpt_obj <- .find_obj(fit$fit, is_timegpt_fit, scan_depth)
+  if (is.null(timegpt_obj) && is_timegpt_fit(fit$fit)) timegpt_obj <- fit$fit
+
+  if (!is.null(timegpt_obj)) {
+    # Training data stats - external regressors
+    if (!is.null(timegpt_obj$train_data)) {
+      train_data <- timegpt_obj$train_data
+      original_cols <- colnames(train_data)[grepl("_original", colnames(train_data))]
+      if (length(original_cols) > 0) {
+        clean_names <- gsub("_original$", "", original_cols) # Remove _original suffix
+        eng_tbl <- dplyr::bind_rows(eng_tbl, .kv("engine_param", "n_external_regressors", as.character(length(clean_names))))
+        eng_tbl <- dplyr::bind_rows(eng_tbl, .kv("engine_param", "external_regressors", paste(clean_names, collapse = ", ")))
+      } else {
+        eng_tbl <- dplyr::bind_rows(eng_tbl, .kv("engine_param", "n_external_regressors", "0"))
+      }
+    }
+
+    # API Details
+    api_key_set <- !is.null(getOption("NIXTLA_API_KEY")) || nzchar(Sys.getenv("NIXTLA_API_KEY"))
+    base_url <- getOption("NIXTLA_BASE_URL") %||% Sys.getenv("NIXTLA_BASE_URL")
+    if (api_key_set) {
+      if (!is.null(base_url) && nzchar(base_url)) {
+        is_azure <- grepl("^https?://[^/]*azure", base_url, ignore.case = TRUE)
+        eng_tbl <- dplyr::bind_rows(eng_tbl, .kv("engine_param", "api_endpoint_type", ifelse(is_azure, "azure", "nixtla_default")))
+      } else {
+        eng_tbl <- dplyr::bind_rows(eng_tbl, .kv("engine_param", "api_endpoint_type", "nixtla_default"))
+      }
+    } else {
+      eng_tbl <- dplyr::bind_rows(eng_tbl, .kv("engine_param", "api_configuration", "failed"))
+    }
+
+    # Model variant (long-horizon flag)
+    if (!is.null(timegpt_obj$forecast_horizon) && !is.null(timegpt_obj$frequency)) {
+      date_type <- get_date_type(timegpt_obj$frequency)
+      if (!is.null(date_type) && is_long_horizon_forecast(timegpt_obj$forecast_horizon, date_type)) {
+        eng_tbl <- dplyr::bind_rows(eng_tbl, .kv("engine_param", "model_variant", "timegpt-1-long-horizon"))
+      }
+    }
+  }
+
+  .assemble_output(
+    preds_tbl, outs_tbl, steps_tbl, args_tbl, eng_tbl,
+    class(fit$fit)[1], engine,
+    unquote_values = FALSE, digits = digits
+  )
+}
+
+
 #' Convert various R objects to a single character string
 #' @noRd
 .chr1 <- function(x) {
diff --git a/R/train_models.R b/R/train_models.R
index 7ff4854..ba1c477 100644
--- a/R/train_models.R
+++ b/R/train_models.R
@@ -141,6 +141,8 @@ train_models <- function(run_info,
 
   global_model_list <- list_global_models()
   fs_model_list <- list_multivariate_models()
+  # Remove timegpt from feature selection list as it doesn't use traditional features, prevents Combo getting dropped from feature selection
+  fs_model_list <- setdiff(fs_model_list, "timegpt")
 
   if (sum(model_workflow_list %in% global_model_list) == 0 & run_global_models) {
     run_global_models <- FALSE