holmesgpt 0.12.3a1__py3-none-any.whl → 0.12.5__py3-none-any.whl

holmes/__init__.py

@@ -1,5 +1,5 @@
 # This is patched by github actions during release
-__version__ = "0.12.3-alpha.1"
+__version__ = "0.12.5"
 
 # Re-export version functions from version module for backward compatibility
 from .version import (

holmes/config.py

@@ -4,36 +4,40 @@ import os
 import os.path
 from enum import Enum
 from pathlib import Path
-from typing import Any, List, Optional, Union
+from typing import TYPE_CHECKING, Any, List, Optional, Union
 
 import yaml  # type: ignore
 from pydantic import BaseModel, ConfigDict, FilePath, SecretStr
 
 from holmes.common.env_vars import ROBUSTA_AI, ROBUSTA_API_ENDPOINT, ROBUSTA_CONFIG_PATH
-from holmes.core.llm import LLM, DefaultLLM
-from holmes.core.runbooks import RunbookManager
-from holmes.core.supabase_dal import SupabaseDal
-from holmes.core.tool_calling_llm import IssueInvestigator, ToolCallingLLM
 from holmes.core.tools_utils.tool_executor import ToolExecutor
 from holmes.core.toolset_manager import ToolsetManager
-from holmes.plugins.destinations.slack import SlackDestination
 from holmes.plugins.runbooks import (
     RunbookCatalog,
     load_builtin_runbooks,
     load_runbook_catalog,
     load_runbooks_from_file,
 )
-from holmes.plugins.sources.github import GitHubSource
-from holmes.plugins.sources.jira import JiraServiceManagementSource, JiraSource
-from holmes.plugins.sources.opsgenie import OpsGenieSource
-from holmes.plugins.sources.pagerduty import PagerDutySource
-from holmes.plugins.sources.prometheus.plugin import AlertManagerSource
+
+# Source plugin imports moved to their respective create methods to speed up startup
+if TYPE_CHECKING:
+    from holmes.core.llm import LLM
+    from holmes.core.supabase_dal import SupabaseDal
+    from holmes.core.tool_calling_llm import IssueInvestigator, ToolCallingLLM
+    from holmes.plugins.destinations.slack import SlackDestination
+    from holmes.plugins.sources.github import GitHubSource
+    from holmes.plugins.sources.jira import JiraServiceManagementSource, JiraSource
+    from holmes.plugins.sources.opsgenie import OpsGenieSource
+    from holmes.plugins.sources.pagerduty import PagerDutySource
+    from holmes.plugins.sources.prometheus.plugin import AlertManagerSource
+
+from holmes.core.config import config_path_dir
 from holmes.utils.definitions import RobustaConfig
 from holmes.utils.env import replace_env_vars_values
 from holmes.utils.file_utils import load_yaml_file
 from holmes.utils.pydantic_utils import RobustaBaseConfig, load_model_from_file
 
-DEFAULT_CONFIG_LOCATION = os.path.expanduser("~/.holmes/config.yaml")
+DEFAULT_CONFIG_LOCATION = os.path.join(config_path_dir, "config.yaml")
 MODEL_LIST_FILE_LOCATION = os.environ.get(
     "MODEL_LIST_FILE_LOCATION", "/etc/holmes/config/model_list.yaml"
 )
@@ -111,6 +115,7 @@ class Config(RobustaBaseConfig):
     should_try_robusta_ai: bool = False  # if True, we will try to load the Robusta AI model, in cli we aren't trying to load it.
 
     toolsets: Optional[dict[str, dict[str, Any]]] = None
+    mcp_servers: Optional[dict[str, dict[str, Any]]] = None
 
     _server_tool_executor: Optional[ToolExecutor] = None
 
@@ -121,6 +126,7 @@ class Config(RobustaBaseConfig):
         if not self._toolset_manager:
             self._toolset_manager = ToolsetManager(
                 toolsets=self.toolsets,
+                mcp_servers=self.mcp_servers,
                 custom_toolsets=self.custom_toolsets,
                 custom_toolsets_from_cli=self.custom_toolsets_from_cli,
             )
@@ -246,7 +252,7 @@ class Config(RobustaBaseConfig):
         return runbook_catalog
 
     def create_console_tool_executor(
-        self, dal: Optional[SupabaseDal], refresh_status: bool = False
+        self, dal: Optional["SupabaseDal"], refresh_status: bool = False
     ) -> ToolExecutor:
         """
         Creates a ToolExecutor instance configured for CLI usage. This executor manages the available tools
@@ -262,7 +268,7 @@ class Config(RobustaBaseConfig):
         )
         return ToolExecutor(cli_toolsets)
 
-    def create_tool_executor(self, dal: Optional[SupabaseDal]) -> ToolExecutor:
+    def create_tool_executor(self, dal: Optional["SupabaseDal"]) -> ToolExecutor:
         """
         Creates ToolExecutor for the server endpoints
         """
@@ -282,53 +288,73 @@ class Config(RobustaBaseConfig):
 
     def create_console_toolcalling_llm(
         self,
-        dal: Optional[SupabaseDal] = None,
+        dal: Optional["SupabaseDal"] = None,
         refresh_toolsets: bool = False,
         tracer=None,
-    ) -> ToolCallingLLM:
+    ) -> "ToolCallingLLM":
         tool_executor = self.create_console_tool_executor(dal, refresh_toolsets)
+        from holmes.core.tool_calling_llm import ToolCallingLLM
+
         return ToolCallingLLM(
             tool_executor, self.max_steps, self._get_llm(tracer=tracer)
         )
 
     def create_toolcalling_llm(
         self,
-        dal: Optional[SupabaseDal] = None,
+        dal: Optional["SupabaseDal"] = None,
         model: Optional[str] = None,
         tracer=None,
-    ) -> ToolCallingLLM:
+    ) -> "ToolCallingLLM":
         tool_executor = self.create_tool_executor(dal)
+        from holmes.core.tool_calling_llm import ToolCallingLLM
+
         return ToolCallingLLM(
             tool_executor, self.max_steps, self._get_llm(model, tracer)
         )
 
     def create_issue_investigator(
         self,
-        dal: Optional[SupabaseDal] = None,
+        dal: Optional["SupabaseDal"] = None,
         model: Optional[str] = None,
         tracer=None,
-    ) -> IssueInvestigator:
+    ) -> "IssueInvestigator":
         all_runbooks = load_builtin_runbooks()
         for runbook_path in self.custom_runbooks:
             all_runbooks.extend(load_runbooks_from_file(runbook_path))
 
+        from holmes.core.runbooks import RunbookManager
+
         runbook_manager = RunbookManager(all_runbooks)
         tool_executor = self.create_tool_executor(dal)
+        from holmes.core.tool_calling_llm import IssueInvestigator
+
         return IssueInvestigator(
-            tool_executor, runbook_manager, self.max_steps, self._get_llm(model, tracer)
+            tool_executor=tool_executor,
+            runbook_manager=runbook_manager,
+            max_steps=self.max_steps,
+            llm=self._get_llm(model, tracer),
+            cluster_name=self.cluster_name,
         )
 
     def create_console_issue_investigator(
-        self, dal: Optional[SupabaseDal] = None
-    ) -> IssueInvestigator:
+        self, dal: Optional["SupabaseDal"] = None
+    ) -> "IssueInvestigator":
         all_runbooks = load_builtin_runbooks()
         for runbook_path in self.custom_runbooks:
             all_runbooks.extend(load_runbooks_from_file(runbook_path))
 
+        from holmes.core.runbooks import RunbookManager
+
         runbook_manager = RunbookManager(all_runbooks)
         tool_executor = self.create_console_tool_executor(dal=dal)
+        from holmes.core.tool_calling_llm import IssueInvestigator
+
         return IssueInvestigator(
-            tool_executor, runbook_manager, self.max_steps, self._get_llm()
+            tool_executor=tool_executor,
+            runbook_manager=runbook_manager,
+            max_steps=self.max_steps,
+            llm=self._get_llm(),
+            cluster_name=self.cluster_name,
         )
 
     def validate_jira_config(self):
@@ -343,7 +369,9 @@ class Config(RobustaBaseConfig):
         if self.jira_api_key is None:
             raise ValueError("--jira-api-key must be specified")
 
-    def create_jira_source(self) -> JiraSource:
+    def create_jira_source(self) -> "JiraSource":
+        from holmes.plugins.sources.jira import JiraSource
+
         self.validate_jira_config()
 
         return JiraSource(
@@ -353,7 +381,9 @@ class Config(RobustaBaseConfig):
             jql_query=self.jira_query,  # type: ignore
         )
 
-    def create_jira_service_management_source(self) -> JiraServiceManagementSource:
+    def create_jira_service_management_source(self) -> "JiraServiceManagementSource":
+        from holmes.plugins.sources.jira import JiraServiceManagementSource
+
         self.validate_jira_config()
 
         return JiraServiceManagementSource(
@@ -363,7 +393,9 @@ class Config(RobustaBaseConfig):
             jql_query=self.jira_query,  # type: ignore
         )
 
-    def create_github_source(self) -> GitHubSource:
+    def create_github_source(self) -> "GitHubSource":
+        from holmes.plugins.sources.github import GitHubSource
+
         if not self.github_url or not (
             self.github_url.startswith("http://")
             or self.github_url.startswith("https://")
@@ -384,7 +416,9 @@ class Config(RobustaBaseConfig):
             query=self.github_query,
         )
 
-    def create_pagerduty_source(self) -> PagerDutySource:
+    def create_pagerduty_source(self) -> "PagerDutySource":
+        from holmes.plugins.sources.pagerduty import PagerDutySource
+
         if self.pagerduty_api_key is None:
             raise ValueError("--pagerduty-api-key must be specified")
 
@@ -394,7 +428,9 @@ class Config(RobustaBaseConfig):
             incident_key=self.pagerduty_incident_key,
         )
 
-    def create_opsgenie_source(self) -> OpsGenieSource:
+    def create_opsgenie_source(self) -> "OpsGenieSource":
+        from holmes.plugins.sources.opsgenie import OpsGenieSource
+
         if self.opsgenie_api_key is None:
             raise ValueError("--opsgenie-api-key must be specified")
 
@@ -408,7 +444,9 @@ class Config(RobustaBaseConfig):
             ),
         )
 
-    def create_alertmanager_source(self) -> AlertManagerSource:
+    def create_alertmanager_source(self) -> "AlertManagerSource":
+        from holmes.plugins.sources.prometheus.plugin import AlertManagerSource
+
         return AlertManagerSource(
             url=self.alertmanager_url,  # type: ignore
             username=self.alertmanager_username,
@@ -417,14 +455,16 @@ class Config(RobustaBaseConfig):
             filepath=self.alertmanager_file,
         )
 
-    def create_slack_destination(self):
+    def create_slack_destination(self) -> "SlackDestination":
+        from holmes.plugins.destinations.slack import SlackDestination
+
         if self.slack_token is None:
             raise ValueError("--slack-token must be specified")
         if self.slack_channel is None:
             raise ValueError("--slack-channel must be specified")
         return SlackDestination(self.slack_token.get_secret_value(), self.slack_channel)
 
-    def _get_llm(self, model_key: Optional[str] = None, tracer=None) -> LLM:
+    def _get_llm(self, model_key: Optional[str] = None, tracer=None) -> "LLM":
         api_key = self.api_key.get_secret_value() if self.api_key else None
         model = self.model
         model_params = {}
@@ -438,6 +478,8 @@ class Config(RobustaBaseConfig):
             api_key = model_params.pop("api_key", api_key)
             model = model_params.pop("model", model)
 
+        from holmes.core.llm import DefaultLLM
+
         return DefaultLLM(model, api_key, model_params, tracer)  # type: ignore
 
     def get_models_list(self) -> List[str]:
@@ -450,7 +492,7 @@ class Config(RobustaBaseConfig):
 class TicketSource(BaseModel):
     config: Config
     output_instructions: list[str]
-    source: Union[JiraServiceManagementSource, PagerDutySource]
+    source: Union["JiraServiceManagementSource", "PagerDutySource"]
 
     model_config = ConfigDict(arbitrary_types_allowed=True)
 

holmes/core/config.py

@@ -0,0 +1,5 @@
+import os
+
+config_path_dir: str = os.environ.get(
+    "HOLMES_CONFIGPATH_DIR", os.path.expanduser("~/.holmes")
+)

holmes/core/conversations.py

@@ -2,6 +2,7 @@ from typing import Dict, List, Optional
 
 import sentry_sdk
 
+from holmes.config import Config
 from holmes.core.models import (
     ToolCallConversationResult,
     IssueChatRequest,
@@ -60,6 +61,7 @@ def truncate_tool_messages(conversation_history: list, tool_size: int) -> None:
 def build_issue_chat_messages(
     issue_chat_request: IssueChatRequest,
     ai: ToolCallingLLM,
+    config: Config,
     global_instructions: Optional[Instructions] = None,
 ):
     """
@@ -130,6 +132,7 @@ def build_issue_chat_messages(
                     "tools_called_for_investigation": tools_for_investigation,
                     "issue": issue_chat_request.issue_type,
                     "toolsets": ai.tool_executor.toolsets,
+                    "cluster_name": config.cluster_name,
                 },
             )
             messages = [
@@ -149,6 +152,7 @@ def build_issue_chat_messages(
             "tools_called_for_investigation": None,
             "issue": issue_chat_request.issue_type,
             "toolsets": ai.tool_executor.toolsets,
+            "cluster_name": config.cluster_name,
         }
         system_prompt_without_tools = load_and_render_prompt(
             template_path, template_context_without_tools
@@ -181,6 +185,7 @@ def build_issue_chat_messages(
             "tools_called_for_investigation": truncated_investigation_result_tool_calls,
             "issue": issue_chat_request.issue_type,
             "toolsets": ai.tool_executor.toolsets,
+            "cluster_name": config.cluster_name,
         }
         system_prompt_with_truncated_tools = load_and_render_prompt(
             template_path, truncated_template_context
@@ -221,6 +226,7 @@ def build_issue_chat_messages(
         "tools_called_for_investigation": None,
         "issue": issue_chat_request.issue_type,
         "toolsets": ai.tool_executor.toolsets,
+        "cluster_name": config.cluster_name,
     }
     system_prompt_without_tools = load_and_render_prompt(
         template_path, template_context_without_tools
@@ -243,6 +249,7 @@ def build_issue_chat_messages(
         "tools_called_for_investigation": truncated_investigation_result_tool_calls,
         "issue": issue_chat_request.issue_type,
         "toolsets": ai.tool_executor.toolsets,
+        "cluster_name": config.cluster_name,
     }
     system_prompt_with_truncated_tools = load_and_render_prompt(
         template_path, template_context
@@ -255,7 +262,7 @@ def build_issue_chat_messages(
 
 
 def add_or_update_system_prompt(
-    conversation_history: List[Dict[str, str]], ai: ToolCallingLLM
+    conversation_history: List[Dict[str, str]], ai: ToolCallingLLM, config: Config
 ):
     """Either add the system prompt or replace an existing system prompt.
     As a 'defensive' measure, this code will only replace an existing system prompt if it is the
@@ -266,6 +273,7 @@ def add_or_update_system_prompt(
     template_path = "builtin://generic_ask_conversation.jinja2"
     context = {
         "toolsets": ai.tool_executor.toolsets,
+        "cluster_name": config.cluster_name,
     }
 
     system_prompt = load_and_render_prompt(template_path, context)
@@ -293,6 +301,7 @@ def build_chat_messages(
     ask: str,
     conversation_history: Optional[List[Dict[str, str]]],
     ai: ToolCallingLLM,
+    config: Config,
     global_instructions: Optional[Instructions] = None,
 ) -> List[dict]:
     """
@@ -349,7 +358,7 @@ def build_chat_messages(
         conversation_history = conversation_history.copy()
 
     conversation_history = add_or_update_system_prompt(
-        conversation_history=conversation_history, ai=ai
+        conversation_history=conversation_history, ai=ai, config=config
     )
 
     ask = add_global_instructions_to_user_prompt(ask, global_instructions)
@@ -382,6 +391,7 @@ def build_chat_messages(
 def build_workload_health_chat_messages(
     workload_health_chat_request: WorkloadHealthChatRequest,
     ai: ToolCallingLLM,
+    config: Config,
     global_instructions: Optional[Instructions] = None,
 ):
     """
@@ -454,6 +464,7 @@ def build_workload_health_chat_messages(
                     "tools_called_for_workload": tools_for_workload,
                     "resource": resource,
                     "toolsets": ai.tool_executor.toolsets,
+                    "cluster_name": config.cluster_name,
                 },
             )
             messages = [
@@ -473,6 +484,7 @@ def build_workload_health_chat_messages(
             "tools_called_for_workload": None,
             "resource": resource,
             "toolsets": ai.tool_executor.toolsets,
+            "cluster_name": config.cluster_name,
         }
         system_prompt_without_tools = load_and_render_prompt(
             template_path, template_context_without_tools
@@ -505,6 +517,7 @@ def build_workload_health_chat_messages(
             "tools_called_for_workload": truncated_workload_result_tool_calls,
             "resource": resource,
             "toolsets": ai.tool_executor.toolsets,
+            "cluster_name": config.cluster_name,
         }
         system_prompt_with_truncated_tools = load_and_render_prompt(
             template_path, truncated_template_context
@@ -545,6 +558,7 @@ def build_workload_health_chat_messages(
         "tools_called_for_workload": None,
         "resource": resource,
         "toolsets": ai.tool_executor.toolsets,
+        "cluster_name": config.cluster_name,
     }
     system_prompt_without_tools = load_and_render_prompt(
         template_path, template_context_without_tools
@@ -567,6 +581,7 @@ def build_workload_health_chat_messages(
         "tools_called_for_workload": truncated_workload_result_tool_calls,
         "resource": resource,
         "toolsets": ai.tool_executor.toolsets,
+        "cluster_name": config.cluster_name,
     }
     system_prompt_with_truncated_tools = load_and_render_prompt(
         template_path, template_context

holmes/core/investigation.py

@@ -132,6 +132,7 @@ def get_investigation_context(
             "sections": sections,
             "structured_output": request_structured_output_from_llm,
             "toolsets": ai.tool_executor.toolsets,
+            "cluster_name": config.cluster_name,
         },
     )
 

holmes/core/llm.py

@@ -12,7 +12,6 @@ import litellm
 import os
 from holmes.common.env_vars import (
     THINKING,
-    TEMPERATURE,
 )
 
 
@@ -218,6 +217,7 @@ class DefaultLLM(LLM):
         if self.args.get("thinking", None):
             litellm.modify_params = True
 
+        self.args.setdefault("temperature", temperature)
         # Get the litellm module to use (wrapped or unwrapped)
         litellm_to_use = self.tracer.wrap_llm(litellm) if self.tracer else litellm
 
@@ -225,7 +225,6 @@ class DefaultLLM(LLM):
             model=self.model,
             api_key=self.api_key,
             messages=messages,
-            temperature=temperature or self.args.pop("temperature", TEMPERATURE),
             response_format=response_format,
             drop_params=drop_params,
             stream=stream,

holmes/core/prompt.py

@@ -1,11 +1,13 @@
 from rich.console import Console
-from typing import Optional, List, Dict
+from typing import Optional, List, Dict, Any, Union
 from pathlib import Path
+from holmes.plugins.prompts import load_and_render_prompt
+from holmes.plugins.runbooks import RunbookCatalog
 
 
 def append_file_to_user_prompt(user_prompt: str, file_path: Path) -> str:
     with file_path.open("r") as f:
-        user_prompt += f"\n\n<attached-file path='{file_path.absolute()}>'\n{f.read()}\n</attached-file>"
+        user_prompt += f"\n\n<attached-file path='{file_path.absolute()}'>\n{f.read()}\n</attached-file>"
 
     return user_prompt
 
@@ -25,11 +27,34 @@ def append_all_files_to_user_prompt(
 
 def build_initial_ask_messages(
     console: Console,
-    system_prompt_rendered: str,
     initial_user_prompt: str,
     file_paths: Optional[List[Path]],
+    tool_executor: Any,  # ToolExecutor type
+    runbooks: Union[RunbookCatalog, Dict, None] = None,
+    system_prompt_additions: Optional[str] = None,
 ) -> List[Dict]:
-    """Build the initial messages for the AI call."""
+    """Build the initial messages for the AI call.
+
+    Args:
+        console: Rich console for output
+        initial_user_prompt: The user's prompt
+        file_paths: Optional list of files to include
+        tool_executor: The tool executor with available toolsets
+        runbooks: Optional runbook catalog
+        system_prompt_additions: Optional additional system prompt content
+    """
+    # Load and render system prompt internally
+    system_prompt_template = "builtin://generic_ask.jinja2"
+    template_context = {
+        "toolsets": tool_executor.toolsets,
+        "runbooks": runbooks or {},
+        "system_prompt_additions": system_prompt_additions or "",
+    }
+    system_prompt_rendered = load_and_render_prompt(
+        system_prompt_template, template_context
+    )
+
+    # Append files to user prompt
     user_prompt_with_files = append_all_files_to_user_prompt(
         console, initial_user_prompt, file_paths
     )

holmes/core/supabase_dal.py

@@ -7,6 +7,7 @@ import threading
 from datetime import datetime, timedelta
 from typing import Dict, List, Optional, Tuple
 from uuid import uuid4
+import gzip
 
 import yaml  # type: ignore
 from cachetools import TTLCache  # type: ignore
@@ -291,6 +292,52 @@ class SupabaseDal:
 
         return changes_data
 
+    def unzip_evidence_file(self, data):
+        try:
+            evidence_list = json.loads(data.get("data", "[]"))
+            if not evidence_list:
+                return data
+
+            evidence = evidence_list[0]
+            raw_data = evidence.get("data")
+
+            if evidence.get("type") != "gz" or not raw_data:
+                return data
+
+            # Strip "b'...'" or 'b"..."' markers if present
+            if raw_data.startswith("b'") and raw_data.endswith("'"):
+                raw_data = raw_data[2:-1]
+            elif raw_data.startswith('b"') and raw_data.endswith('"'):
+                raw_data = raw_data[2:-1]
+
+            gz_bytes = base64.b64decode(raw_data)
+            decompressed = gzip.decompress(gz_bytes).decode("utf-8")
+
+            evidence["data"] = decompressed
+            data["data"] = json.dumps([evidence])
+            return data
+
+        except Exception:
+            logging.exception(f"Unknown issue unzipping gz finding: {data}")
+            return data
+
+    def extract_relevant_issues(self, evidence):
+        enrichment_blacklist = {"text_file", "graph", "ai_analysis", "holmes"}
+        data = [
+            enrich
+            for enrich in evidence.data
+            if enrich.get("enrichment_type") not in enrichment_blacklist
+        ]
+
+        unzipped_files = [
+            self.unzip_evidence_file(enrich)
+            for enrich in evidence.data
+            if enrich.get("enrichment_type") == "text_file"
+        ]
+
+        data.extend(unzipped_files)
+        return data
+
     def get_issue_data(self, issue_id: Optional[str]) -> Optional[Dict]:
         # TODO this could be done in a single atomic SELECT, but there is no
         # foreign key relation between Issues and Evidence.
@@ -320,12 +367,7 @@ class SupabaseDal:
             .filter("issue_id", "eq", issue_id)
             .execute()
         )
-        enrichment_blacklist = {"text_file", "graph", "ai_analysis", "holmes"}
-        data = [
-            enrich
-            for enrich in evidence.data
-            if enrich.get("enrichment_type") not in enrichment_blacklist
-        ]
+        data = self.extract_relevant_issues(evidence)
 
         issue_data["evidence"] = data
 
@@ -470,13 +512,7 @@ class SupabaseDal:
                 .execute()
             )
 
-            enrichment_blacklist = {"text_file", "graph", "ai_analysis", "holmes"}
-            data = [
-                evidence.get("data")
-                for evidence in res.data
-                if evidence.get("enrichment_type") not in enrichment_blacklist
-            ]
-            return data
+            return self.extract_relevant_issues(res)
 
         except Exception:
             logging.exception("failed to fetch workload issues data", exc_info=True)

holmes/core/tool_calling_llm.py

@@ -15,7 +15,11 @@ from pydantic import BaseModel
 from pydantic_core import from_json
 from rich.console import Console
 
-from holmes.common.env_vars import ROBUSTA_API_ENDPOINT, STREAM_CHUNKS_PER_PARSE
+from holmes.common.env_vars import (
+    ROBUSTA_API_ENDPOINT,
+    STREAM_CHUNKS_PER_PARSE,
+    TEMPERATURE,
+)
 from holmes.core.investigation_structured_output import (
     DEFAULT_SECTIONS,
     REQUEST_STRUCTURED_OUTPUT_FROM_LLM,
@@ -40,6 +44,7 @@ from holmes.utils.global_instructions import (
 from holmes.utils.tags import format_tags_in_string, parse_messages_tags
 from holmes.core.tools_utils.tool_executor import ToolExecutor
 from holmes.core.tracing import DummySpan
+from holmes.utils.colors import AI_COLOR
 
 
 def format_tool_result_data(tool_result: StructuredToolResult) -> str:
@@ -285,6 +290,7 @@ class ToolCallingLLM:
                     messages=parse_messages_tags(messages),
                     tools=tools,
                     tool_choice=tool_choice,
+                    temperature=TEMPERATURE,
                     response_format=response_format,
                     drop_params=True,
                 )
@@ -328,6 +334,15 @@ class ToolCallingLLM:
 
             tools_to_call = getattr(response_message, "tool_calls", None)
             text_response = response_message.content
+
+            if (
+                hasattr(response_message, "reasoning_content")
+                and response_message.reasoning_content
+            ):
+                logging.debug(
+                    f"[bold {AI_COLOR}]AI (reasoning) 🤔:[/bold {AI_COLOR}] {response_message.reasoning_content}\n"
+                )
+
             if not tools_to_call:
                 # For chatty models post process and summarize the result
                 # this only works for calls where user prompt is explicitly passed through
@@ -357,6 +372,11 @@ class ToolCallingLLM:
                     messages=messages,
                 )
 
+            if text_response and text_response.strip():
+                logging.info(f"[bold {AI_COLOR}]AI:[/bold {AI_COLOR}] {text_response}")
+            logging.info(
+                f"The AI requested [bold]{len(tools_to_call) if tools_to_call else 0}[/bold] tool call(s)."
+            )
             perf_timing.measure("pre-tool-calls")
             with concurrent.futures.ThreadPoolExecutor(max_workers=16) as executor:
                 futures = []
@@ -610,6 +630,7 @@ class ToolCallingLLM:
                             "messages": parse_messages_tags(messages),  # type: ignore
                             "tools": tools,
                             "tool_choice": tool_choice,
+                            "temperature": TEMPERATURE,
                             "response_format": response_format,
                             "stream": True,
                             "drop_param": True,
@@ -634,6 +655,7 @@ class ToolCallingLLM:
                         messages=parse_messages_tags(messages),  # type: ignore
                         tools=tools,
                         tool_choice=tool_choice,
+                        temperature=TEMPERATURE,
                         response_format=response_format,
                         stream=False,
                         drop_params=True,
@@ -745,9 +767,11 @@ class IssueInvestigator(ToolCallingLLM):
         runbook_manager: RunbookManager,
         max_steps: int,
         llm: LLM,
+        cluster_name: Optional[str],
     ):
         super().__init__(tool_executor, max_steps, llm)
         self.runbook_manager = runbook_manager
+        self.cluster_name = cluster_name
 
     def investigate(
         self,
@@ -806,6 +830,7 @@ class IssueInvestigator(ToolCallingLLM):
                 "sections": sections,
                 "structured_output": request_structured_output_from_llm,
                 "toolsets": self.tool_executor.toolsets,
+                "cluster_name": self.cluster_name,
             },
         )