holmesgpt 0.13.2__py3-none-any.whl → 0.16.2a0__py3-none-any.whl

holmes/plugins/toolsets/grafana/trace_parser.py

@@ -187,7 +187,7 @@ def format_traces_list(trace_data: Dict) -> str:
                 else "\n"
             )
             trace_str += f"\tstartTime={unix_nano_to_rfc3339(int(trace.get('startTimeUnixNano')))}"
-            trace_str += f" rootServiceName={trace.get('trootServiceName')}"
+            trace_str += f" rootServiceName={trace.get('rootServiceName')}"
             trace_str += f" rootTraceName={trace.get('rootTraceName')}"
             traces_str.append(trace_str)
         return "\n".join(traces_str)

holmes/plugins/toolsets/internet/internet.py

@@ -6,6 +6,7 @@ from typing import Any, Optional, Tuple, Dict, List
 from requests import RequestException, Timeout  # type: ignore
 from holmes.core.tools import (
     Tool,
+    ToolInvokeContext,
     ToolParameter,
     Toolset,
     ToolsetTag,
@@ -15,7 +16,7 @@ from markdownify import markdownify
 from bs4 import BeautifulSoup
 
 import requests  # type: ignore
-from holmes.core.tools import StructuredToolResult, ToolResultStatus
+from holmes.core.tools import StructuredToolResult, StructuredToolResultStatus
 from holmes.plugins.toolsets.utils import toolset_name_for_one_liner
 
 
@@ -25,7 +26,7 @@ INTERNET_TOOLSET_USER_AGENT = os.environ.get(
     "Mozilla/5.0 (X11; Linux x86_64; rv:128.0; holmesgpt;) Gecko/20100101 Firefox/128.0",
 )
 INTERNET_TOOLSET_TIMEOUT_SECONDS = int(
-    os.environ.get("INTERNET_TOOLSET_TIMEOUT_SECONDS", "60")
+    os.environ.get("INTERNET_TOOLSET_TIMEOUT_SECONDS", "5")
 )
 
 SELECTORS_TO_REMOVE = [
@@ -186,9 +187,7 @@ class FetchWebpage(Tool):
             toolset=toolset,  # type: ignore
         )
 
-    def _invoke(
-        self, params: dict, user_approved: bool = False
-    ) -> StructuredToolResult:
+    def _invoke(self, params: dict, context: ToolInvokeContext) -> StructuredToolResult:
         url: str = params["url"]
 
         additional_headers = (
@@ -199,7 +198,7 @@ class FetchWebpage(Tool):
         if not content:
             logging.error(f"Failed to retrieve content from {url}")
             return StructuredToolResult(
-                status=ToolResultStatus.ERROR,
+                status=StructuredToolResultStatus.ERROR,
                 error=f"Failed to retrieve content from {url}",
                 params=params,
             )
@@ -211,7 +210,7 @@ class FetchWebpage(Tool):
             content = html_to_markdown(content)
 
         return StructuredToolResult(
-            status=ToolResultStatus.SUCCESS,
+            status=StructuredToolResultStatus.SUCCESS,
             data=content,
             params=params,
         )

holmes/plugins/toolsets/internet/notion.py

@@ -4,6 +4,7 @@ import json
 from typing import Any, Dict, Tuple
 from holmes.core.tools import (
     Tool,
+    ToolInvokeContext,
     ToolParameter,
     ToolsetTag,
 )
@@ -13,7 +14,7 @@ from holmes.plugins.toolsets.internet.internet import (
 )
 from holmes.core.tools import (
     StructuredToolResult,
-    ToolResultStatus,
+    StructuredToolResultStatus,
 )
 from holmes.plugins.toolsets.utils import toolset_name_for_one_liner
 
@@ -44,9 +45,7 @@ class FetchNotion(Tool):
             return f"https://api.notion.com/v1/blocks/{notion_id}/children"
         return url  # Return original URL if no match is found
 
-    def _invoke(
-        self, params: dict, user_approved: bool = False
-    ) -> StructuredToolResult:
+    def _invoke(self, params: dict, context: ToolInvokeContext) -> StructuredToolResult:
         url: str = params["url"]
 
         # Get headers from the toolset configuration
@@ -59,13 +58,13 @@ class FetchNotion(Tool):
         if not content:
             logging.error(f"Failed to retrieve content from {url}")
             return StructuredToolResult(
-                status=ToolResultStatus.ERROR,
+                status=StructuredToolResultStatus.ERROR,
                 error=f"Failed to retrieve content from {url}",
                 params=params,
             )
 
         return StructuredToolResult(
-            status=ToolResultStatus.SUCCESS,
+            status=StructuredToolResultStatus.SUCCESS,
             data=self.parse_notion_content(content),
             params=params,
         )

holmes/plugins/toolsets/investigator/core_investigation.py

@@ -3,20 +3,39 @@ import os
 from typing import Any, Dict
 
 from uuid import uuid4
+
 from holmes.core.todo_tasks_formatter import format_tasks
 from holmes.core.tools import (
+    StructuredToolResult,
+    StructuredToolResultStatus,
+    Tool,
+    ToolInvokeContext,
+    ToolParameter,
     Toolset,
     ToolsetTag,
-    ToolParameter,
-    Tool,
-    StructuredToolResult,
-    ToolResultStatus,
 )
 from holmes.plugins.toolsets.investigator.model import Task, TaskStatus
 
+TODO_WRITE_TOOL_NAME = "TodoWrite"
+
+
+def parse_tasks(todos_data: Any) -> list[Task]:
+    tasks = []
+
+    for todo_item in todos_data:
+        if isinstance(todo_item, dict):
+            task = Task(
+                id=todo_item.get("id", str(uuid4())),
+                content=todo_item.get("content", ""),
+                status=TaskStatus(todo_item.get("status", "pending")),
+            )
+            tasks.append(task)
+
+    return tasks
+
 
 class TodoWriteTool(Tool):
-    name: str = "TodoWrite"
+    name: str = TODO_WRITE_TOOL_NAME
     description: str = "Save investigation tasks to break down complex problems into manageable sub-tasks. ALWAYS provide the COMPLETE list of all tasks, not just the ones being updated."
     parameters: Dict[str, ToolParameter] = {
         "todos": ToolParameter(
@@ -28,7 +47,11 @@ class TodoWriteTool(Tool):
                 properties={
                     "id": ToolParameter(type="string", required=True),
                     "content": ToolParameter(type="string", required=True),
-                    "status": ToolParameter(type="string", required=True),
+                    "status": ToolParameter(
+                        type="string",
+                        required=True,
+                        enum=["pending", "in_progress", "completed"],
+                    ),
                 },
             ),
         ),
@@ -57,41 +80,28 @@ class TodoWriteTool(Tool):
         content_width = max(max_content_width, len("Content"))
         status_width = max(max_status_display_width, len("Status"))
 
-        # Build table
         separator = f"+{'-' * (id_width + 2)}+{'-' * (content_width + 2)}+{'-' * (status_width + 2)}+"
         header = f"| {'ID':<{id_width}} | {'Content':<{content_width}} | {'Status':<{status_width}} |"
-
-        # Log the table
-        logging.info("Updated Investigation Tasks:")
-        logging.info(separator)
-        logging.info(header)
-        logging.info(separator)
+        tasks_to_display = []
 
         for task in tasks:
             status_display = f"{status_icons[task.status.value]} {task.status.value}"
             row = f"| {task.id:<{id_width}} | {task.content:<{content_width}} | {status_display:<{status_width}} |"
-            logging.info(row)
+            tasks_to_display.append(row)
 
-        logging.info(separator)
+        logging.info(
+            f"Task List:\n{separator}\n{header}\n{separator}\n"
+            + "\n".join(tasks_to_display)
+            + f"\n{separator}"
+        )
 
-    def _invoke(
-        self, params: dict, user_approved: bool = False
-    ) -> StructuredToolResult:
+    def _invoke(self, params: dict, context: ToolInvokeContext) -> StructuredToolResult:
         try:
             todos_data = params.get("todos", [])
 
-            tasks = []
-
-            for todo_item in todos_data:
-                if isinstance(todo_item, dict):
-                    task = Task(
-                        id=todo_item.get("id", str(uuid4())),
-                        content=todo_item.get("content", ""),
-                        status=TaskStatus(todo_item.get("status", "pending")),
-                    )
-                    tasks.append(task)
+            tasks = parse_tasks(todos_data=todos_data)
 
-            logging.info(f"Tasks: {len(tasks)}")
+            logging.debug(f"Tasks: {len(tasks)}")
 
             self.print_tasks_table(tasks)
             formatted_tasks = format_tasks(tasks)
@@ -103,7 +113,7 @@ class TodoWriteTool(Tool):
                 response_data += "No tasks currently in the investigation plan."
 
             return StructuredToolResult(
-                status=ToolResultStatus.SUCCESS,
+                status=StructuredToolResultStatus.SUCCESS,
                 data=response_data,
                 params=params,
             )
@@ -111,14 +121,13 @@ class TodoWriteTool(Tool):
         except Exception as e:
             logging.exception("error using todowrite tool")
             return StructuredToolResult(
-                status=ToolResultStatus.ERROR,
+                status=StructuredToolResultStatus.ERROR,
                 error=f"Failed to process tasks: {str(e)}",
                 params=params,
             )
 
     def get_parameterized_one_liner(self, params: Dict) -> str:
-        todos = params.get("todos", [])
-        return f"Write {todos} investigation tasks"
+        return "Update investigation tasks"
 
 
 class CoreInvestigationToolset(Toolset):
@@ -133,7 +142,6 @@ class CoreInvestigationToolset(Toolset):
             tags=[ToolsetTag.CORE],
             is_default=True,
         )
-        logging.info("Core investigation toolset loaded")
 
     def get_example_config(self) -> Dict[str, Any]:
         return {}

holmes/plugins/toolsets/kafka.py

@@ -27,8 +27,9 @@ from holmes.core.tools import (
     CallablePrerequisite,
     StructuredToolResult,
     Tool,
+    ToolInvokeContext,
     ToolParameter,
-    ToolResultStatus,
+    StructuredToolResultStatus,
     Toolset,
     ToolsetTag,
 )
@@ -153,15 +154,13 @@ class ListKafkaConsumers(BaseKafkaTool):
             toolset=toolset,
         )
 
-    def _invoke(
-        self, params: dict, user_approved: bool = False
-    ) -> StructuredToolResult:
+    def _invoke(self, params: dict, context: ToolInvokeContext) -> StructuredToolResult:
         try:
             kafka_cluster_name = get_param_or_raise(params, "kafka_cluster_name")
             client = self.get_kafka_client(kafka_cluster_name)
             if client is None:
                 return StructuredToolResult(
-                    status=ToolResultStatus.ERROR,
+                    status=StructuredToolResultStatus.ERROR,
                     error="No admin_client on toolset. This toolset is misconfigured.",
                     params=params,
                 )
@@ -190,7 +189,7 @@ class ListKafkaConsumers(BaseKafkaTool):
             if errors_text:
                 result_text = result_text + "\n\n" + errors_text
             return StructuredToolResult(
-                status=ToolResultStatus.SUCCESS,
+                status=StructuredToolResultStatus.SUCCESS,
                 data=result_text,
                 params=params,
             )
@@ -198,7 +197,7 @@ class ListKafkaConsumers(BaseKafkaTool):
             error_msg = f"Failed to list consumer groups: {str(e)}"
             logging.error(error_msg)
             return StructuredToolResult(
-                status=ToolResultStatus.ERROR,
+                status=StructuredToolResultStatus.ERROR,
                 error=error_msg,
                 params=params,
             )
@@ -228,16 +227,14 @@ class DescribeConsumerGroup(BaseKafkaTool):
             toolset=toolset,
         )
 
-    def _invoke(
-        self, params: dict, user_approved: bool = False
-    ) -> StructuredToolResult:
+    def _invoke(self, params: dict, context: ToolInvokeContext) -> StructuredToolResult:
         group_id = params["group_id"]
         try:
             kafka_cluster_name = get_param_or_raise(params, "kafka_cluster_name")
             client = self.get_kafka_client(kafka_cluster_name)
             if client is None:
                 return StructuredToolResult(
-                    status=ToolResultStatus.ERROR,
+                    status=StructuredToolResultStatus.ERROR,
                     error="No admin_client on toolset. This toolset is misconfigured.",
                     params=params,
                 )
@@ -247,13 +244,13 @@ class DescribeConsumerGroup(BaseKafkaTool):
             if futures.get(group_id):
                 group_metadata = futures.get(group_id).result()
                 return StructuredToolResult(
-                    status=ToolResultStatus.SUCCESS,
+                    status=StructuredToolResultStatus.SUCCESS,
                     data=yaml.dump(convert_to_dict(group_metadata)),
                     params=params,
                 )
             else:
                 return StructuredToolResult(
-                    status=ToolResultStatus.ERROR,
+                    status=StructuredToolResultStatus.ERROR,
                     error="Group not found",
                     params=params,
                 )
@@ -261,7 +258,7 @@ class DescribeConsumerGroup(BaseKafkaTool):
             error_msg = f"Failed to describe consumer group {group_id}: {str(e)}"
             logging.error(error_msg)
             return StructuredToolResult(
-                status=ToolResultStatus.ERROR,
+                status=StructuredToolResultStatus.ERROR,
                 error=error_msg,
                 params=params,
             )
@@ -286,22 +283,20 @@ class ListTopics(BaseKafkaTool):
             toolset=toolset,
         )
 
-    def _invoke(
-        self, params: dict, user_approved: bool = False
-    ) -> StructuredToolResult:
+    def _invoke(self, params: dict, context: ToolInvokeContext) -> StructuredToolResult:
         try:
             kafka_cluster_name = get_param_or_raise(params, "kafka_cluster_name")
             client = self.get_kafka_client(kafka_cluster_name)
             if client is None:
                 return StructuredToolResult(
-                    status=ToolResultStatus.ERROR,
+                    status=StructuredToolResultStatus.ERROR,
                     error="No admin_client on toolset. This toolset is misconfigured.",
                     params=params,
                 )
 
             topics = client.list_topics()
             return StructuredToolResult(
-                status=ToolResultStatus.SUCCESS,
+                status=StructuredToolResultStatus.SUCCESS,
                 data=yaml.dump(convert_to_dict(topics)),
                 params=params,
             )
@@ -309,7 +304,7 @@ class ListTopics(BaseKafkaTool):
             error_msg = f"Failed to list topics: {str(e)}"
             logging.error(error_msg)
             return StructuredToolResult(
-                status=ToolResultStatus.ERROR,
+                status=StructuredToolResultStatus.ERROR,
                 error=error_msg,
                 params=params,
             )
@@ -344,16 +339,14 @@ class DescribeTopic(BaseKafkaTool):
             toolset=toolset,
         )
 
-    def _invoke(
-        self, params: dict, user_approved: bool = False
-    ) -> StructuredToolResult:
+    def _invoke(self, params: dict, context: ToolInvokeContext) -> StructuredToolResult:
         topic_name = params["topic_name"]
         try:
             kafka_cluster_name = get_param_or_raise(params, "kafka_cluster_name")
             client = self.get_kafka_client(kafka_cluster_name)
             if client is None:
                 return StructuredToolResult(
-                    status=ToolResultStatus.ERROR,
+                    status=StructuredToolResultStatus.ERROR,
                     error="No admin_client on toolset. This toolset is misconfigured.",
                     params=params,
                 )
@@ -373,7 +366,7 @@ class DescribeTopic(BaseKafkaTool):
                 result["configuration"] = convert_to_dict(config)
 
             return StructuredToolResult(
-                status=ToolResultStatus.SUCCESS,
+                status=StructuredToolResultStatus.SUCCESS,
                 data=yaml.dump(result),
                 params=params,
             )
@@ -381,7 +374,7 @@ class DescribeTopic(BaseKafkaTool):
             error_msg = f"Failed to describe topic {topic_name}: {str(e)}"
             logging.error(error_msg, exc_info=True)
             return StructuredToolResult(
-                status=ToolResultStatus.ERROR,
+                status=StructuredToolResultStatus.ERROR,
                 error=error_msg,
                 params=params,
             )
@@ -469,16 +462,14 @@ class FindConsumerGroupsByTopic(BaseKafkaTool):
             toolset=toolset,
         )
 
-    def _invoke(
-        self, params: dict, user_approved: bool = False
-    ) -> StructuredToolResult:
+    def _invoke(self, params: dict, context: ToolInvokeContext) -> StructuredToolResult:
         topic_name = params["topic_name"]
         try:
             kafka_cluster_name = get_param_or_raise(params, "kafka_cluster_name")
             client = self.get_kafka_client(kafka_cluster_name)
             if client is None:
                 return StructuredToolResult(
-                    status=ToolResultStatus.ERROR,
+                    status=StructuredToolResultStatus.ERROR,
                     error="No admin_client on toolset. This toolset is misconfigured.",
                     params=params,
                 )
@@ -530,7 +521,7 @@ class FindConsumerGroupsByTopic(BaseKafkaTool):
                 result_text = result_text + "\n\n" + errors_text
 
             return StructuredToolResult(
-                status=ToolResultStatus.SUCCESS,
+                status=StructuredToolResultStatus.SUCCESS,
                 data=result_text,
                 params=params,
             )
@@ -540,7 +531,7 @@ class FindConsumerGroupsByTopic(BaseKafkaTool):
             )
             logging.error(error_msg)
             return StructuredToolResult(
-                status=ToolResultStatus.ERROR,
+                status=StructuredToolResultStatus.ERROR,
                 error=error_msg,
                 params=params,
             )
@@ -559,12 +550,10 @@ class ListKafkaClusters(BaseKafkaTool):
             toolset=toolset,
         )
 
-    def _invoke(
-        self, params: dict, user_approved: bool = False
-    ) -> StructuredToolResult:
+    def _invoke(self, params: dict, context: ToolInvokeContext) -> StructuredToolResult:
         cluster_names = list(self.toolset.clients.keys())
         return StructuredToolResult(
-            status=ToolResultStatus.SUCCESS,
+            status=StructuredToolResultStatus.SUCCESS,
             data="Available Kafka Clusters:\n" + "\n".join(cluster_names),
             params=params,
         )

holmes/plugins/toolsets/kubernetes.yaml

@@ -8,6 +8,10 @@ toolsets:
     prerequisites:
       - command: "kubectl version --client"
 
+    # Note: Many tools in this toolset use transformers with llm_summarize
+    # to automatically summarize large kubectl outputs when a fast model is configured.
+    # This reduces context window usage while preserving key information for debugging.
+
     tools:
       - name: "kubectl_describe"
         description: >
@@ -17,6 +21,20 @@ toolsets:
             - 'describe pod xyz-123'
             - 'show service xyz-123 in namespace my-ns'
         command: "kubectl describe {{ kind }} {{ name }}{% if namespace %} -n {{ namespace }}{% endif %}"
+        transformers:
+          - name: llm_summarize
+            config:
+              input_threshold: 1000
+              prompt: |
+                Summarize this kubectl describe output focusing on:
+                - What needs attention or immediate action
+                - Resource status and health indicators
+                - Any errors, warnings, or non-standard states
+                - Key configuration details that could affect functionality
+                - When possible, mention exact field names so the user can grep for specific details
+                - Be concise: aim for ≤ 50% of the original length; avoid repeating defaults/healthy/unchanged details
+                - Prefer aggregates and counts; list only outliers and actionable items
+                - Keep grep-friendly: include exact field names/values that matter
 
       - name: "kubectl_get_by_name"
         description: "Run `kubectl get <kind> <name> --show-labels`"
@@ -25,10 +43,36 @@ toolsets:
       - name: "kubectl_get_by_kind_in_namespace"
         description: "Run `kubectl get <kind> -n <namespace> --show-labels` to get all resources of a given type in namespace"
         command: "kubectl get --show-labels -o wide {{ kind }} -n {{namespace}}"
+        transformers:
+          - name: llm_summarize
+            config:
+              input_threshold: 1000
+              prompt: |
+                Summarize this kubectl output focusing on:
+                - What needs attention or immediate action
+                - Group similar resources into aggregate descriptions
+                - Make sure to mention outliers, errors, and non-standard states
+                - List healthy resources as aggregate descriptions
+                - When listing unhealthy resources, also try to use aggregate descriptions when possible
+                - When possible, mention exact keywords so the user can rerun the command with | grep <keyword> and drill down
+                - Be concise and avoid expansion: target ≤ 50% of input size; prefer counts + outliers over full listings
 
       - name: "kubectl_get_by_kind_in_cluster"
         description: "Run `kubectl get -A <kind> --show-labels` to get all resources of a given type in the cluster"
         command: "kubectl get -A --show-labels -o wide {{ kind }}"
+        transformers:
+          - name: llm_summarize
+            config:
+              input_threshold: 1000
+              prompt: |
+                Summarize this kubectl output focusing on:
+                - What needs attention or immediate action
+                - Group similar resources into a single line and description
+                - Make sure to mention outliers, errors, and non-standard states
+                - List healthy resources as aggregate descriptions
+                - When listing unhealthy resources, also try to use aggregate descriptions when possible
+                - When possible, mention exact keywords so the user can rerun the command with | grep <keyword> and drill down on the parts they care about
+                - Strive for ≤ 50% of the original size; keep results compact and grep-friendly (one line per aggregate)
 
       - name: "kubectl_find_resource"
         description: "Run `kubectl get {{ kind }} -A --show-labels | grep {{ keyword }}` to find a resource where you know a substring of the name, IP, namespace, or labels"
@@ -42,95 +86,25 @@ toolsets:
         description: "Retrieve the events for a specific Kubernetes resource. `resource_type` can be any kubernetes resource type: 'pod', 'service', 'deployment', 'job', 'node', etc."
         command: "kubectl events --for {{resource_type}}/{{ resource_name }}{% if namespace %} -n {{ namespace }}{% endif %}"
 
-      - name: "kubectl_memory_requests_all_namespaces"
-        description: "Fetch and display memory requests for all pods across all namespaces in MiB, summing requests across multiple containers where applicable and handling binary, decimal, and millibyte units correctly."
-        command: |
-          kubectl get pods --all-namespaces -o custom-columns="NAMESPACE:.metadata.namespace,NAME:.metadata.name,MEMORY_REQUEST:.spec.containers[*].resources.requests.memory" --no-headers | \
-          awk '
-            function convert_to_mib(value) {
-              if (value ~ /^[0-9]+e[0-9]+$/) return (value + 0) / (1024 * 1024); # Scientific notation
-              if (value ~ /m$/) return (value + 0) / (1024^2 * 1000);           # Millibytes (m)
-              if (value ~ /Ei$/) return (value + 0) * 1024^6 / (1024^2);        # Binary units
-              if (value ~ /Pi$/) return (value + 0) * 1024^5 / (1024^2);
-              if (value ~ /Ti$/) return (value + 0) * 1024^4 / (1024^2);
-              if (value ~ /Gi$/) return (value + 0) * 1024^3 / (1024^2);
-              if (value ~ /Mi$/) return (value + 0);
-              if (value ~ /Ki$/) return (value + 0) / 1024;
-              if (value ~ /E$/) return (value + 0) * 1000^6 / (1024^2);         # Decimal units
-              if (value ~ /P$/) return (value + 0) * 1000^5 / (1024^2);
-              if (value ~ /T$/) return (value + 0) * 1000^4 / (1024^2);
-              if (value ~ /G$/) return (value + 0) * 1000^3 / (1024^2);
-              if (value ~ /M$/) return (value + 0) * 1000^2 / (1024^2);
-              if (value ~ /k$/) return (value + 0) * 1000 / (1024^2);
-              return (value + 0) / (1024 * 1024);                               # Default: bytes
-            }
-            function sum_memory(requests) {
-              gsub(/^[ \t]+|[ \t]+$/, "", requests);
-              if (requests == "" || requests == "<none>") return 0;
-              split(requests, arr, ",");
-              total = 0;
-              for (i in arr) {
-                if (arr[i] != "<none>") total += convert_to_mib(arr[i]);
-              }
-              return total;
-            }
-            {
-              namespace = $1;
-              name = $2;
-              requests = $3;
-              for (i=4; i<=NF; i++) {
-                requests = requests " " $i;
-              }
-              print namespace, name, sum_memory(requests) " Mi";
-            }' | sort -k3 -nr
-
-      - name: "kubectl_memory_requests_namespace"
-        description: "Fetch and display memory requests for all pods in a specified namespace in MiB, summing requests across multiple containers where applicable and handling binary, decimal, and millibyte units correctly."
-        command: |
-          kubectl get pods -n {{ namespace }} -o custom-columns="NAMESPACE:.metadata.namespace,NAME:.metadata.name,MEMORY_REQUEST:.spec.containers[*].resources.requests.memory" --no-headers | \
-          awk '
-            function convert_to_mib(value) {
-              if (value ~ /^[0-9]+e[0-9]+$/) return (value + 0) / (1024 * 1024); # Scientific notation
-              if (value ~ /m$/) return (value + 0) / (1024^2 * 1000);           # Millibytes (m)
-              if (value ~ /Ei$/) return (value + 0) * 1024^6 / (1024^2);        # Binary units
-              if (value ~ /Pi$/) return (value + 0) * 1024^5 / (1024^2);
-              if (value ~ /Ti$/) return (value + 0) * 1024^4 / (1024^2);
-              if (value ~ /Gi$/) return (value + 0) * 1024^3 / (1024^2);
-              if (value ~ /Mi$/) return (value + 0);
-              if (value ~ /Ki$/) return (value + 0) / 1024;
-              if (value ~ /E$/) return (value + 0) * 1000^6 / (1024^2);         # Decimal units
-              if (value ~ /P$/) return (value + 0) * 1000^5 / (1024^2);
-              if (value ~ /T$/) return (value + 0) * 1000^4 / (1024^2);
-              if (value ~ /G$/) return (value + 0) * 1000^3 / (1024^2);
-              if (value ~ /M$/) return (value + 0) * 1000^2 / (1024^2);
-              if (value ~ /k$/) return (value + 0) * 1000 / (1024^2);
-              return (value + 0) / (1024 * 1024);                               # Default: bytes
-            }
-            function sum_memory(requests) {
-              gsub(/^[ \t]+|[ \t]+$/, "", requests);
-              if (requests == "" || requests == "<none>") return 0;
-              split(requests, arr, ",");
-              total = 0;
-              for (i in arr) {
-                if (arr[i] != "<none>") total += convert_to_mib(arr[i]);
-              }
-              return total;
-            }
-            {
-              namespace = $1;
-              name = $2;
-              requests = $3;
-              for (i=4; i<=NF; i++) {
-                requests = requests " " $i;
-              }
-              print namespace, name, sum_memory(requests) " Mi";
-            }' | sort -k3 -nr
-
       - name: "kubernetes_jq_query"
         user_description: "Query Kubernetes Resources: kubectl get {{kind}} --all-namespaces -o json | jq -r {{jq_expr}}"
         description: >
           Use kubectl to get json for all resources of a specific kind pipe the results to jq to filter them. Do not worry about escaping the jq_expr it will be done by the system on an unescaped expression that you give. e.g. give an expression like .items[] | .spec.containers[].image | select(test("^gcr.io/") | not)
         command: kubectl get {{ kind }} --all-namespaces -o json | jq -r {{ jq_expr }}
+        transformers:
+          - name: llm_summarize
+            config:
+              input_threshold: 1000
+              prompt: |
+                Summarize this jq query output focusing on:
+                - Key patterns and commonalities in the data
+                - Notable outliers, anomalies, or items that need attention
+                - Group similar results into aggregate descriptions when possible
+                - Highlight any empty results, null values, or missing data
+                - When applicable, mention specific resource names, namespaces, or values that stand out
+                - Organize findings in a structured way that helps with troubleshooting
+                - Be concise: aim for ≤ 50% of the original text; prioritize aggregates and actionable outliers
+                - Include grep-ready keys/values; avoid repeating entire objects or unchanged defaults
 
       - name: "kubernetes_count"
         user_description: "Count Kubernetes Resources: kubectl get {{kind}} --all-namespaces -o json | jq -c -r {{ jq_expr }}"