Skip to content

OpenRAG is a comprehensive, single package Retrieval-Augmented Generation platform built on Langflow, Docling, and Opensearch.

Python

fix: make openai not be required for ingestion to work (#1101)

Lucas Oliveirabaca26fb32f6
diff --git a/docker-compose.yml b/docker-compose.yml
index 0aef3a172..68e4206ba 100644
--- a/docker-compose.yml
+++ b/docker-compose.yml
@@ -133,7 +133,7 @@ services:
- LANGFLOW_DEACTIVATE_TRACING
- OPENAI_API_KEY=${OPENAI_API_KEY}
- ANTHROPIC_API_KEY=${ANTHROPIC_API_KEY}
- - WATSONX_API_KEY=${WATSONX_API_KEY}
+ - WATSONX_APIKEY=${WATSONX_API_KEY}
- WATSONX_URL=${WATSONX_URL:-${WATSONX_ENDPOINT}}
- WATSONX_PROJECT_ID=${WATSONX_PROJECT_ID}
- OLLAMA_BASE_URL=${OLLAMA_ENDPOINT}
@@ -160,7 +160,7 @@ services:
- MIMETYPE=None
- FILESIZE=0
- SELECTED_EMBEDDING_MODEL=${SELECTED_EMBEDDING_MODEL:-text-embedding-3-small}
- - LANGFLOW_VARIABLES_TO_GET_FROM_ENVIRONMENT=JWT,OPENRAG-QUERY-FILTER,OPENSEARCH_PASSWORD,OPENSEARCH_URL,DOCLING_SERVE_URL,OWNER,OWNER_NAME,OWNER_EMAIL,CONNECTOR_TYPE,DOCUMENT_ID,SOURCE_URL,ALLOWED_USERS,ALLOWED_GROUPS,FILENAME,MIMETYPE,FILESIZE,SELECTED_EMBEDDING_MODEL,OPENAI_API_KEY,ANTHROPIC_API_KEY,WATSONX_API_KEY,WATSONX_URL,WATSONX_PROJECT_ID,OLLAMA_BASE_URL,OPENSEARCH_INDEX_NAME
+ - LANGFLOW_VARIABLES_TO_GET_FROM_ENVIRONMENT=JWT,OPENRAG-QUERY-FILTER,OPENSEARCH_PASSWORD,OPENSEARCH_URL,DOCLING_SERVE_URL,OWNER,OWNER_NAME,OWNER_EMAIL,CONNECTOR_TYPE,DOCUMENT_ID,SOURCE_URL,ALLOWED_USERS,ALLOWED_GROUPS,FILENAME,MIMETYPE,FILESIZE,SELECTED_EMBEDDING_MODEL,OPENAI_API_KEY,ANTHROPIC_API_KEY,WATSONX_APIKEY,WATSONX_URL,WATSONX_PROJECT_ID,OLLAMA_BASE_URL,OPENSEARCH_INDEX_NAME
- LANGFLOW_LOG_LEVEL=DEBUG
- LANGFLOW_WORKERS=${LANGFLOW_WORKERS:-1}
- LANGFLOW_AUTO_LOGIN=${LANGFLOW_AUTO_LOGIN}
diff --git a/flows/ingestion_flow.json b/flows/ingestion_flow.json
index f389252bd..f62f63fba 100644
--- a/flows/ingestion_flow.json
+++ b/flows/ingestion_flow.json
@@ -4579,7 +4579,7 @@
],
"list": false,
"list_add_label": "Add More",
- "load_from_db": true,
+ "load_from_db": false,
"name": "project_id",
"override_skip": false,
"placeholder": "",
@@ -4591,7 +4591,7 @@
"trace_as_metadata": true,
"track_in_telemetry": false,
"type": "str",
- "value": "WATSONX_PROJECT_ID"
+ "value": ""
},
"request_timeout": {
"_input_type": "FloatInput",
diff --git a/flows/openrag_agent.json b/flows/openrag_agent.json
index 320d6789b..8a33b996f 100644
--- a/flows/openrag_agent.json
+++ b/flows/openrag_agent.json
@@ -4748,7 +4748,7 @@
],
"list": false,
"list_add_label": "Add More",
- "load_from_db": true,
+ "load_from_db": false,
"name": "project_id",
"override_skip": false,
"placeholder": "",
@@ -4760,7 +4760,7 @@
"trace_as_metadata": true,
"track_in_telemetry": false,
"type": "str",
- "value": "WATSONX_PROJECT_ID"
+ "value": ""
},
"request_timeout": {
"_input_type": "FloatInput",
diff --git a/flows/openrag_nudges.json b/flows/openrag_nudges.json
index 7fbc7bc4e..eae87db31 100644
--- a/flows/openrag_nudges.json
+++ b/flows/openrag_nudges.json
@@ -1236,7 +1236,7 @@
"dynamic": false,
"info": "Falls back to OPENAI_API_KEY environment variable",
"input_types": [],
- "load_from_db": true,
+ "load_from_db": false,
"name": "api_key",
"override_skip": false,
"password": true,
@@ -1247,7 +1247,7 @@
"title_case": false,
"track_in_telemetry": false,
"type": "str",
- "value": "OPENAI_API_KEY"
+ "value": ""
},
"base_url_ibm_watsonx": {
"_input_type": "DropdownInput",
@@ -3987,7 +3987,7 @@
],
"list": false,
"list_add_label": "Add More",
- "load_from_db": true,
+ "load_from_db": false,
"name": "project_id",
"override_skip": false,
"placeholder": "",
@@ -3999,7 +3999,7 @@
"trace_as_metadata": true,
"track_in_telemetry": false,
"type": "str",
- "value": "WATSONX_PROJECT_ID"
+ "value": ""
},
"request_timeout": {
"_input_type": "FloatInput",
diff --git a/flows/openrag_url_mcp.json b/flows/openrag_url_mcp.json
index d2235b2a5..5050fa044 100644
--- a/flows/openrag_url_mcp.json
+++ b/flows/openrag_url_mcp.json
@@ -5738,7 +5738,7 @@
],
"list": false,
"list_add_label": "Add More",
- "load_from_db": true,
+ "load_from_db": false,
"name": "project_id",
"override_skip": false,
"placeholder": "",
@@ -5750,7 +5750,7 @@
"trace_as_metadata": true,
"track_in_telemetry": false,
"type": "str",
- "value": "WATSONX_PROJECT_ID"
+ "value": ""
},
"request_timeout": {
"_input_type": "FloatInput",
diff --git a/kubernetes/helm/openrag/templates/langflow/deployment.yaml b/kubernetes/helm/openrag/templates/langflow/deployment.yaml
index 058fcc815..aa57b8762 100644
--- a/kubernetes/helm/openrag/templates/langflow/deployment.yaml
+++ b/kubernetes/helm/openrag/templates/langflow/deployment.yaml
@@ -251,7 +251,7 @@ spec:
{{- end }}
{{- if .Values.llmProviders.watsonx.enabled }}
- - name: WATSONX_API_KEY
+ - name: WATSONX_APIKEY
valueFrom:
secretKeyRef:
name: {{ include "openrag.fullname" . }}-llm-providers
@@ -267,7 +267,7 @@ spec:
name: {{ include "openrag.fullname" . }}-llm-providers
key: watsonx-project-id
{{- else }}
- - name: WATSONX_API_KEY
+ - name: WATSONX_APIKEY
value: "None"
- name: WATSONX_URL
value: "None"
diff --git a/kubernetes/helm/openrag/values.yaml b/kubernetes/helm/openrag/values.yaml
index 97d42e781..1eb7fff48 100644
--- a/kubernetes/helm/openrag/values.yaml
+++ b/kubernetes/helm/openrag/values.yaml
@@ -123,7 +123,7 @@ langflow:
host: "https://cloud.langfuse.com"
# Variables to expose to flows
- variablesToGetFromEnvironment: "JWT,OPENRAG-QUERY-FILTER,OPENSEARCH_PASSWORD,OPENSEARCH_URL,OPENSEARCH_INDEX_NAME,DOCLING_SERVE_URL,OWNER,OWNER_NAME,OWNER_EMAIL,CONNECTOR_TYPE,DOCUMENT_ID,SOURCE_URL,ALLOWED_USERS,ALLOWED_GROUPS,FILENAME,MIMETYPE,FILESIZE,SELECTED_EMBEDDING_MODEL,OPENAI_API_KEY,ANTHROPIC_API_KEY,WATSONX_API_KEY,WATSONX_URL,WATSONX_PROJECT_ID,OLLAMA_BASE_URL"
+ variablesToGetFromEnvironment: "JWT,OPENRAG-QUERY-FILTER,OPENSEARCH_PASSWORD,OPENSEARCH_URL,OPENSEARCH_INDEX_NAME,DOCLING_SERVE_URL,OWNER,OWNER_NAME,OWNER_EMAIL,CONNECTOR_TYPE,DOCUMENT_ID,SOURCE_URL,ALLOWED_USERS,ALLOWED_GROUPS,FILENAME,MIMETYPE,FILESIZE,SELECTED_EMBEDDING_MODEL,OPENAI_API_KEY,ANTHROPIC_API_KEY,WATSONX_APIKEY,WATSONX_URL,WATSONX_PROJECT_ID,OLLAMA_BASE_URL"
# Probes
livenessProbe:
diff --git a/src/api/settings.py b/src/api/settings.py
index c89695311..eb6bd83a3 100644
--- a/src/api/settings.py
+++ b/src/api/settings.py
@@ -1299,9 +1299,9 @@ async def _update_langflow_global_variables(config):
# WatsonX global variables
if config.providers.watsonx.api_key:
await clients._create_langflow_global_variable(
- "WATSONX_API_KEY", config.providers.watsonx.api_key, modify=True
+ "WATSONX_APIKEY", config.providers.watsonx.api_key, modify=True
)
- logger.info("Set WATSONX_API_KEY global variable in Langflow")
+ logger.info("Set WATSONX_APIKEY global variable in Langflow")
if config.providers.watsonx.project_id:
await clients._create_langflow_global_variable(
@@ -1396,29 +1396,45 @@ async def _update_mcp_servers_with_provider_credentials(config, session_manager
async def _update_langflow_model_values(config, flows_service):
- """Update model values across Langflow flows"""
+ """Update model values across Langflow flows for all configured providers"""
try:
- # Update LLM model values
+ # 1. Update ONLY the current LLM provider
llm_provider = config.agent.llm_provider.lower()
-
await flows_service.change_langflow_model_value(
llm_provider,
llm_model=config.agent.llm_model,
+ force_llm_update=True
)
logger.info(
f"Successfully updated Langflow flows for LLM provider {llm_provider}"
)
- # Update embedding model values
- embedding_provider = config.knowledge.embedding_provider.lower()
+ # 2. Update ALL configured embedding providers
+ embedding_providers = []
+ if config.providers.openai.configured:
+ embedding_providers.append("openai")
+ if config.providers.watsonx.configured:
+ embedding_providers.append("watsonx")
+ if config.providers.ollama.configured:
+ embedding_providers.append("ollama")
+
+ current_embedding_provider = config.knowledge.embedding_provider.lower()
+ for provider in embedding_providers:
+ # Use configured model for current provider, or None (first available) for others
+ embedding_model = (
+ config.knowledge.embedding_model
+ if provider == current_embedding_provider
+ else None
+ )
- await flows_service.change_langflow_model_value(
- embedding_provider,
- embedding_model=config.knowledge.embedding_model,
- )
- logger.info(
- f"Successfully updated Langflow flows for embedding provider {embedding_provider}"
- )
+ await flows_service.change_langflow_model_value(
+ provider,
+ embedding_model=embedding_model,
+ force_embedding_update=True
+ )
+ logger.info(
+ f"Successfully updated Langflow flows for embedding provider {provider}"
+ )
except Exception as e:
logger.error(f"Failed to update Langflow model values: {str(e)}")
diff --git a/src/services/flows_service.py b/src/services/flows_service.py
index d770b47fc..a0644be18 100644
--- a/src/services/flows_service.py
+++ b/src/services/flows_service.py
@@ -401,71 +401,31 @@ class FlowsService:
f"Updating {flow_type} flow with current configuration settings"
)
- # Get LLM provider (used for most flows)
- llm_provider = config.agent.llm_provider.lower()
- embedding_provider = config.knowledge.embedding_provider.lower()
-
- # Step 2: Update model values for the specific flow being reset
- single_flow_config = [
- {
- "name": flow_type,
- "flow_id": flow_id,
- }
- ]
+ # Update current LLM provider
+ update_result = await self._update_provider_components(
+ {"name": flow_type, "flow_id": flow_id},
+ config.agent.llm_provider.lower(),
+ llm_model=config.agent.llm_model,
+ force_llm_update=True
+ )
- logger.info(f"Updating {flow_type} flow model values")
-
- # For retrieval flow: need to update both LLM and embedding (potentially different providers)
- # For ingest flows: only update embedding
- # For other flows: only update LLM
-
- if flow_type == "retrieval":
- # Retrieval flow uses both LLM and embedding models
- # Update LLM first
- llm_result = await self.change_langflow_model_value(
- provider=llm_provider,
- embedding_model=None,
- llm_model=config.agent.llm_model,
- flow_configs=single_flow_config,
- )
- if not llm_result.get("success"):
- logger.warning(
- f"Failed to update LLM in {flow_type} flow: {llm_result.get('error', 'Unknown error')}"
- )
-
- # Update embedding model
- embedding_result = await self.change_langflow_model_value(
- provider=embedding_provider,
- embedding_model=config.knowledge.embedding_model,
- llm_model=None,
- flow_configs=single_flow_config,
- )
- if not embedding_result.get("success"):
- logger.warning(
- f"Failed to update embedding in {flow_type} flow: {embedding_result.get('error', 'Unknown error')}"
- )
-
- # Consider it successful if either update succeeded
- update_result = {
- "success": llm_result.get("success") or embedding_result.get("success"),
- "llm_result": llm_result,
- "embedding_result": embedding_result,
- }
- elif flow_type in ["ingest", "url_ingest"]:
- # Ingest flows only need embedding model
- update_result = await self.change_langflow_model_value(
- provider=embedding_provider,
- embedding_model=config.knowledge.embedding_model,
- llm_model=None,
- flow_configs=single_flow_config,
- )
- else:
- # Other flows (nudges) only need LLM model
- update_result = await self.change_langflow_model_value(
- provider=llm_provider,
- embedding_model=None,
- llm_model=config.agent.llm_model,
- flow_configs=single_flow_config,
+ # Update all configured embedding providers
+ embedding_providers = []
+ if config.providers.openai.configured:
+ embedding_providers.append("openai")
+ if config.providers.watsonx.configured:
+ embedding_providers.append("watsonx")
+ if config.providers.ollama.configured:
+ embedding_providers.append("ollama")
+
+ current_emb_provider = config.knowledge.embedding_provider.lower()
+ for provider in embedding_providers:
+ model = config.knowledge.embedding_model if provider == current_emb_provider else None
+ await self._update_provider_components(
+ {"name": flow_type, "flow_id": flow_id},
+ provider,
+ embedding_model=model,
+ force_embedding_update=True
)
if update_result.get("success"):
@@ -509,294 +469,6 @@ class FlowsService:
logger.error(f"Error while resetting {flow_type} flow", error=str(e))
return {"success": False, "error": f"Error: {str(e)}"}
- # async def assign_model_provider(self, provider: str, is_embedding: bool = False):
- # """
- # Replace OpenAI components with the specified provider components in all flows
-
- # Args:
- # provider: "watsonx", "ollama", "openai" or "anthropic"
-
- # Returns:
- # dict: Success/error response with details for each flow
- # """
- # if provider not in ["watsonx", "ollama", "openai", "anthropic"]:
- # raise ValueError("provider must be 'watsonx', 'ollama', 'openai', or 'anthropic'")
-
- # if provider == "openai":
- # logger.info("Provider is already OpenAI, no changes needed")
- # return {
- # "success": True,
- # "message": "Provider is already OpenAI, no changes needed",
- # }
-
- # try:
- # # Load component templates based on provider
- # llm_template, embedding_template, llm_text_template = (
- # self._load_component_templates(provider)
- # )
-
- # logger.info(f"Assigning {provider} components")
-
- # # Define flow configurations (removed hardcoded file paths)
- # flow_configs = [
- # {
- # "name": "nudges",
- # "flow_id": NUDGES_FLOW_ID,
- # "embedding_name": OPENAI_EMBEDDING_COMPONENT_DISPLAY_NAME,
- # "llm_text_name": OPENAI_LLM_COMPONENT_DISPLAY_NAME,
- # "llm_name": None,
- # },
- # {
- # "name": "retrieval",
- # "flow_id": LANGFLOW_CHAT_FLOW_ID,
- # "embedding_name": OPENAI_EMBEDDING_COMPONENT_DISPLAY_NAME,
- # "llm_name": OPENAI_LLM_COMPONENT_DISPLAY_NAME,
- # "llm_text_name": None,
- # },
- # {
- # "name": "ingest",
- # "flow_id": LANGFLOW_INGEST_FLOW_ID,
- # "embedding_name": OPENAI_EMBEDDING_COMPONENT_DISPLAY_NAME,
- # "llm_name": None, # Ingestion flow might not have LLM
- # "llm_text_name": None,
- # },
- # {
- # "name": "url_ingest",
- # "flow_id": LANGFLOW_URL_INGEST_FLOW_ID,
- # "embedding_name": OPENAI_EMBEDDING_COMPONENT_DISPLAY_NAME,
- # "llm_name": None,
- # "llm_text_name": None,
- # },
- # ]
-
- # results = []
-
- # # Process each flow sequentially
- # for config in flow_configs:
- # try:
- # result = await self._update_flow_components(
- # config, llm_template, embedding_template, llm_text_template, is_embedding
- # )
- # results.append(result)
- # logger.info(f"Successfully updated {config['name']} flow")
- # except Exception as e:
- # error_msg = f"Failed to update {config['name']} flow: {str(e)}"
- # logger.error(error_msg)
- # results.append(
- # {"flow": config["name"], "success": False, "error": error_msg}
- # )
- # # Continue with other flows even if one fails
-
- # # Check if all flows were successful
- # all_success = all(r.get("success", False) for r in results)
-
- # return {
- # "success": all_success,
- # "message": f"Model provider assignment to {provider} {'completed' if all_success else 'completed with errors'}",
- # "provider": provider,
- # "results": results,
- # }
-
- # except Exception as e:
- # logger.error(f"Error assigning model provider {provider}", error=str(e))
- # return {
- # "success": False,
- # "error": f"Failed to assign model provider: {str(e)}",
- # }
-
- # def _load_component_templates(self, provider: str):
- # """Load component templates for the specified provider"""
- # if provider == "watsonx":
- # llm_path = WATSONX_LLM_COMPONENT_PATH
- # embedding_path = WATSONX_EMBEDDING_COMPONENT_PATH
- # llm_text_path = WATSONX_LLM_TEXT_COMPONENT_PATH
- # elif provider == "ollama":
- # llm_path = OLLAMA_LLM_COMPONENT_PATH
- # embedding_path = OLLAMA_EMBEDDING_COMPONENT_PATH
- # llm_text_path = OLLAMA_LLM_TEXT_COMPONENT_PATH
- # else:
- # raise ValueError(f"Unsupported provider: {provider}")
-
- # # Get the project root directory (same logic as reset_langflow_flow)
- # current_file_dir = os.path.dirname(os.path.abspath(__file__)) # src/services/
- # src_dir = os.path.dirname(current_file_dir) # src/
- # project_root = os.path.dirname(src_dir) # project root
-
- # # Load LLM template
- # llm_full_path = os.path.join(project_root, llm_path)
- # if not os.path.exists(llm_full_path):
- # raise FileNotFoundError(
- # f"LLM component template not found at: {llm_full_path}"
- # )
-
- # with open(llm_full_path, "r") as f:
- # llm_template = json.load(f)
-
- # # Load embedding template
- # embedding_full_path = os.path.join(project_root, embedding_path)
- # if not os.path.exists(embedding_full_path):
- # raise FileNotFoundError(
- # f"Embedding component template not found at: {embedding_full_path}"
- # )
-
- # with open(embedding_full_path, "r") as f:
- # embedding_template = json.load(f)
-
- # # Load LLM Text template
- # llm_text_full_path = os.path.join(project_root, llm_text_path)
- # if not os.path.exists(llm_text_full_path):
- # raise FileNotFoundError(
- # f"LLM Text component template not found at: {llm_text_full_path}"
- # )
-
- # with open(llm_text_full_path, "r") as f:
- # llm_text_template = json.load(f)
-
- # logger.info(f"Loaded component templates for {provider}")
- # return llm_template, embedding_template, llm_text_template
-
- # async def _update_flow_components(
- # self, config, llm_template, embedding_template, llm_text_template, is_embedding: bool = False
- # ):
- # """Update components in a specific flow"""
- # flow_name = config["name"]
- # flow_id = config["flow_id"]
- # old_embedding_name = config["embedding_name"]
- # old_llm_name = config["llm_name"]
- # old_llm_text_name = config["llm_text_name"]
- # # Extract IDs from templates
- # new_llm_id = llm_template["data"]["id"]
- # new_embedding_id = embedding_template["data"]["id"]
- # new_llm_text_id = llm_text_template["data"]["id"]
-
- # # Dynamically find the flow file by ID
- # flow_path = self._find_flow_file_by_id(flow_id)
- # if not flow_path:
- # raise FileNotFoundError(f"Flow file not found for flow ID: {flow_id}")
-
- # # Load flow JSON
- # with open(flow_path, "r") as f:
- # flow_data = json.load(f)
-
- # # Find and replace components
- # components_updated = []
-
- # # Replace embedding component
- # if not DISABLE_INGEST_WITH_LANGFLOW and is_embedding:
- # embedding_node, _ = self._find_node_in_flow(flow_data, display_name=old_embedding_name)
- # if embedding_node:
- # # Preserve position
- # original_position = embedding_node.get("position", {})
-
- # # Replace with new template
- # new_embedding_node = embedding_template.copy()
- # new_embedding_node["position"] = original_position
-
- # # Replace in flow
- # self._replace_node_in_flow(flow_data, old_embedding_name, new_embedding_node)
- # components_updated.append(
- # f"embedding: {old_embedding_name} -> {new_embedding_id}"
- # )
-
- # # Replace LLM component (if exists in this flow)
- # if old_llm_name and not is_embedding:
- # llm_node, _ = self._find_node_in_flow(flow_data, display_name=old_llm_name)
- # if llm_node:
- # # Preserve position
- # original_position = llm_node.get("position", {})
-
- # # Replace with new template
- # new_llm_node = llm_template.copy()
- # new_llm_node["position"] = original_position
-
- # # Replace in flow
- # self._replace_node_in_flow(flow_data, old_llm_name, new_llm_node)
- # components_updated.append(f"llm: {old_llm_name} -> {new_llm_id}")
-
- # # Replace LLM component (if exists in this flow)
- # if old_llm_text_name and not is_embedding:
- # llm_text_node, _ = self._find_node_in_flow(flow_data, display_name=old_llm_text_name)
- # if llm_text_node:
- # # Preserve position
- # original_position = llm_text_node.get("position", {})
-
- # # Replace with new template
- # new_llm_text_node = llm_text_template.copy()
- # new_llm_text_node["position"] = original_position
-
- # # Replace in flow
- # self._replace_node_in_flow(flow_data, old_llm_text_name, new_llm_text_node)
- # components_updated.append(f"llm: {old_llm_text_name} -> {new_llm_text_id}")
-
-
- # old_embedding_id = None
- # old_llm_id = None
- # old_llm_text_id = None
- # if embedding_node:
- # old_embedding_id = embedding_node.get("data", {}).get("id")
- # if old_llm_name and llm_node:
- # old_llm_id = llm_node.get("data", {}).get("id")
- # if old_llm_text_name and llm_text_node:
- # old_llm_text_id = llm_text_node.get("data", {}).get("id")
-
- # # Update all edge references using regex replacement
- # flow_json_str = json.dumps(flow_data)
-
- # # Replace embedding ID references
- # if not DISABLE_INGEST_WITH_LANGFLOW and is_embedding:
- # flow_json_str = re.sub(
- # re.escape(old_embedding_id), new_embedding_id, flow_json_str
- # )
- # flow_json_str = re.sub(
- # re.escape(old_embedding_id.split("-")[0]),
- # new_embedding_id.split("-")[0],
- # flow_json_str,
- # )
-
- # # Replace LLM ID references (if applicable)
- # if old_llm_id and not is_embedding:
- # flow_json_str = re.sub(
- # re.escape(old_llm_id), new_llm_id, flow_json_str
- # )
-
- # flow_json_str = re.sub(
- # re.escape(old_llm_id.split("-")[0]),
- # new_llm_id.split("-")[0],
- # flow_json_str,
- # )
-
- # # Replace text LLM ID references (if applicable)
- # if old_llm_text_id and not is_embedding:
- # flow_json_str = re.sub(
- # re.escape(old_llm_text_id), new_llm_text_id, flow_json_str
- # )
-
- # flow_json_str = re.sub(
- # re.escape(old_llm_text_id.split("-")[0]),
- # new_llm_text_id.split("-")[0],
- # flow_json_str,
- # )
-
- # # Convert back to JSON
- # flow_data = json.loads(flow_json_str)
-
- # # PATCH the updated flow
- # response = await clients.langflow_request(
- # "PATCH", f"/api/v1/flows/{flow_id}", json=flow_data
- # )
-
- # if response.status_code != 200:
- # raise Exception(
- # f"Failed to update flow: HTTP {response.status_code} - {response.text}"
- # )
-
- # return {
- # "flow": flow_name,
- # "success": True,
- # "components_updated": components_updated,
- # "flow_id": flow_id,
- # }
-
def _find_node_in_flow(self, flow_data, node_id=None, display_name=None):
"""
Helper function to find a node in flow data by ID or display name.
@@ -818,6 +490,34 @@ class FlowsService:
return None, None
+ def _find_nodes_in_flow(self, flow_data, display_name=None):
+ """Find all nodes in flow data by display name."""
+ nodes = flow_data.get("data", {}).get("nodes", [])
+ found = []
+ for i, node in enumerate(nodes):
+ node_data = node.get("data", {})
+ node_template = node_data.get("node", {})
+ if display_name and node_template.get("display_name") == display_name:
+ found.append((node, i))
+ return found
+
+ def _get_node_provider(self, node):
+ """Get the provider name currently set in a node."""
+ template = node.get("data", {}).get("node", {}).get("template", {})
+ model_val = template.get("model", {}).get("value")
+ if isinstance(model_val, list) and len(model_val) > 0:
+ return model_val[0].get("provider")
+ return None
+
+ def _get_provider_name_display(self, provider: str):
+ if provider == "watsonx":
+ return "IBM WatsonX"
+ if provider == "ollama":
+ return "Ollama"
+ if provider == "anthropic":
+ return "Anthropic"
+ return "OpenAI"
+
async def _update_flow_field(self, flow_id: str, field_name: str, field_value: str, node_display_name: str = None):
"""
Generic helper function to update any field in any Langflow component.
@@ -1074,12 +774,14 @@ class FlowsService:
logger.info(f"Flow {flow_type} (ID: {flow_id}) does not match reset state")
return reset_flows
-
+
async def change_langflow_model_value(
self,
provider: str,
embedding_model: str = None,
llm_model: str = None,
+ force_embedding_update: bool = False,
+ force_llm_update: bool = False,
flow_configs: list = None,
):
"""
@@ -1089,87 +791,48 @@ class FlowsService:
provider: The provider ("watsonx", "ollama", "openai", "anthropic")
embedding_model: The embedding model name to set
llm_model: The LLM model name to set
- endpoint: The endpoint URL (required for watsonx/ibm provider)
- flow_configs: Optional list of specific flow configs to update. If None, updates all flows.
-
- Returns:
- dict: Success/error response with details for each flow
+ force_embedding_update: If True, update embeddings even if model is None
+ force_llm_update: If True, update LLM even if model is None
+ flow_configs: Optional list of flow configs to update
"""
if provider not in ["watsonx", "ollama", "openai", "anthropic"]:
raise ValueError("provider must be 'watsonx', 'ollama', 'openai', or 'anthropic'")
try:
- logger.info(
- f"Changing dropdown values for provider {provider}, embedding: {embedding_model}, llm: {llm_model}"
- )
-
# Use provided flow_configs or default to all flows
if flow_configs is None:
flow_configs = [
- {
- "name": "nudges",
- "flow_id": NUDGES_FLOW_ID,
- },
- {
- "name": "retrieval",
- "flow_id": LANGFLOW_CHAT_FLOW_ID,
- },
- {
- "name": "ingest",
- "flow_id": LANGFLOW_INGEST_FLOW_ID,
- },
- {
- "name": "url_ingest",
- "flow_id": LANGFLOW_URL_INGEST_FLOW_ID,
- },
+ {"name": "nudges", "flow_id": NUDGES_FLOW_ID},
+ {"name": "retrieval", "flow_id": LANGFLOW_CHAT_FLOW_ID},
+ {"name": "ingest", "flow_id": LANGFLOW_INGEST_FLOW_ID},
+ {"name": "url_ingest", "flow_id": LANGFLOW_URL_INGEST_FLOW_ID},
]
- # Determine target component IDs based on provider
-
results = []
-
- # Process each flow sequentially
for config in flow_configs:
try:
result = await self._update_provider_components(
config,
provider,
- embedding_model,
- llm_model,
+ embedding_model=embedding_model,
+ llm_model=llm_model,
+ force_embedding_update=force_embedding_update,
+ force_llm_update=force_llm_update,
)
results.append(result)
- logger.info(
- f"Successfully updated {config['name']} flow with {provider} models"
- )
except Exception as e:
- error_msg = f"Failed to update {config['name']} flow with {provider} models: {str(e)}"
- logger.error(error_msg)
- results.append(
- {"flow": config["name"], "success": False, "error": error_msg}
- )
- # Continue with other flows even if one fails
-
- # Check if all flows were successful
- all_success = all(r.get("success", False) for r in results)
+ logger.error(f"Error updating {config['name']} flow: {str(e)}")
+ results.append({"flow": config["name"], "success": False, "error": str(e)})
return {
- "success": all_success,
- "message": f"Provider model update {'completed' if all_success else 'completed with errors'}",
- "provider": provider,
- "embedding_model": embedding_model,
- "llm_model": llm_model,
+ "success": all(r.get("success", False) for r in results),
"results": results,
}
-
except Exception as e:
- logger.error(
- f"Error changing provider models for {provider}",
- error=str(e),
- )
- return {
- "success": False,
- "error": f"Failed to change provider models: {str(e)}",
- }
+ logger.error(f"Error in change_langflow_model_value: {str(e)}")
+ return {"success": False, "error": str(e)}
+
+
async def _update_provider_components(
self,
@@ -1177,6 +840,8 @@ class FlowsService:
provider: str,
embedding_model: str = None,
llm_model: str = None,
+ force_embedding_update: bool = False,
+ force_llm_update: bool = False,
):
"""Update provider components and their dropdown values in a flow"""
flow_name = config["name"]
@@ -1195,16 +860,81 @@ class FlowsService:
updates_made = []
# Update embedding component
- if not DISABLE_INGEST_WITH_LANGFLOW and embedding_model:
- embedding_node, _ = self._find_node_in_flow(flow_data, display_name=OPENAI_EMBEDDING_COMPONENT_DISPLAY_NAME)
- if embedding_node:
- if await self._update_component_fields(
- embedding_node, provider, embedding_model
- ):
- updates_made.append(f"embedding model: {embedding_model}")
+ if not DISABLE_INGEST_WITH_LANGFLOW and (embedding_model or force_embedding_update):
+ # Get all embedding nodes in the flow
+ embedding_nodes = self._find_nodes_in_flow(flow_data, display_name=OPENAI_EMBEDDING_COMPONENT_DISPLAY_NAME)
+ logger.info(f"Found {len(embedding_nodes)} embedding nodes in flow {flow_name} with display name '{OPENAI_EMBEDDING_COMPONENT_DISPLAY_NAME}'")
+
+ # Count configured embedding-enabled providers
+ config_obj = get_openrag_config()
+ configured_providers = []
+ if config_obj.providers.openai.configured: configured_providers.append("openai")
+ if config_obj.providers.watsonx.configured: configured_providers.append("watsonx")
+ if config_obj.providers.ollama.configured: configured_providers.append("ollama")
+
+ # Ensure current provider is in the list for counting purposes if it's being configured
+ if provider in ["openai", "watsonx", "ollama"] and provider not in configured_providers:
+ configured_providers.append(provider)
+
+ all_possible = ["openai", "watsonx", "ollama"]
+ configured_providers = [p for p in all_possible if p in configured_providers]
+ provider_count = len(configured_providers)
+ logger.info(f"Configured embedding providers: {configured_providers} (count: {provider_count})")
+
+ # Determine slot mapping context
+ if provider_count == 1:
+ logger.info("Configuration mode: all 3 slots belong to the single active provider")
+ elif provider_count == 2:
+ logger.info("Configuration mode: first 2 slots assigned to providers 1 and 2")
+ elif provider_count == 3:
+ logger.info("Configuration mode: slots 1, 2, and 3 assigned to their respective providers")
+
+ # 1. Check if any node is already this provider - always update those first
+ matched_nodes = []
+ provider_display = self._get_provider_name_display(provider)
+ for node, idx in embedding_nodes:
+ if self._get_node_provider(node) == provider_display:
+ matched_nodes.append((node, idx))
+
+ if matched_nodes:
+ logger.info(f"Found {len(matched_nodes)} nodes already configured for provider '{provider}'")
+ for node, idx in matched_nodes:
+ if await self._update_component_fields(node, provider, embedding_model):
+ updates_made.append(f"embedding model: {embedding_model} (updated existing {provider} node)")
+ else:
+ # 2. No existing node matched, use slot-based logic
+ try:
+ p_index = configured_providers.index(provider)
+ logger.info(f"Using slot-based logic for provider '{provider}' (p_index: {p_index}, total configured: {provider_count})")
+
+ if provider_count == 1:
+ # Single provider mode: update all available nodes (up to 3)
+ logger.info(f"Single provider mode: updating all available embedding nodes (available: {len(embedding_nodes)})")
+ for i in range(min(3, len(embedding_nodes))):
+ node, idx = embedding_nodes[i]
+ if await self._update_component_fields(node, provider, embedding_model):
+ updates_made.append(
+ f"embedding model: {embedding_model} (set node {i+1})"
+ )
+ else:
+ # Multiple providers: each gets one slot based on its list index
+ # This satisfies:
+ # - 2 providers -> node 0 and node 1 updated ("two first nodes")
+ # - 3 providers -> node 0, 1, 2 updated ("each gets its first one")
+ if p_index < len(embedding_nodes):
+ node, idx = embedding_nodes[p_index]
+ logger.info(f"Multiple provider mode: assigning provider '{provider}' to node slot {p_index} (node {p_index+1})")
+ if await self._update_component_fields(node, provider, embedding_model):
+ updates_made.append(
+ f"embedding model: {embedding_model} (set node {p_index+1})"
+ )
+ else:
+ logger.info(f"Provider index {p_index} exceeds available embedding nodes ({len(embedding_nodes)}) - skipping automatic assignment")
+ except ValueError:
+ logger.warning(f"Current provider '{provider}' not found in configured providers list: {configured_providers}")
# Update LLM component (if exists in this flow)
- if llm_model:
+ if llm_model or force_llm_update:
llm_node, _ = self._find_node_in_flow(flow_data, display_name=OPENAI_LLM_COMPONENT_DISPLAY_NAME)
if llm_node:
if await self._update_component_fields(
@@ -1290,14 +1020,76 @@ class FlowsService:
):
"""Update fields in a component node based on provider and component type"""
template = component_node.get("data", {}).get("node", {}).get("template", {})
-
if not template:
return False
updated = False
+ provider_name = self._get_provider_name_display(provider)
+
+ # Enable the model in Langflow first
+ await self._enable_model_in_langflow(provider_name, model_value)
+
+ # Update model field and call custom_component/update endpoint
+ if "model" in template:
+ if "options" not in template["model"]:
+ return False
+
+ # Update template via Langflow API to get latest options
+ template = await self._update_component_langflow(template, template["model"]["value"]) or template
+ component_node["data"]["node"]["template"] = template
+
+ # Find the specific model option for the provider
+ if model_value:
+ model_options = [
+ item for item in template["model"].get("options", [])
+ if item.get("provider") == provider_name and item.get("name") == model_value
+ ]
+ else:
+ # If no specific model provided, pick the first available one for this provider
+ model_options = [
+ item for item in template["model"].get("options", [])
+ if item.get("provider") == provider_name
+ ][:1]
+ if model_options:
+ logger.info(f"Using first available model '{model_options[0].get('name')}' for provider {provider_name}")
+
+ if not model_options:
+ logger.warning(f"Model {model_value or 'ANY'} not found for provider {provider_name}")
+ return False
+
+ template["model"]["value"] = model_options
- provider_name = "IBM WatsonX" if provider == "watsonx" else "Ollama" if provider == "ollama" else "Anthropic" if provider == "anthropic" else "OpenAI"
+ template = await self._update_component_langflow(template, model_options) or template
+ component_node["data"]["node"]["template"] = template
+ updated = True
+
+ # Update provider-specific fields
+ field_mappings = {
+ "api_base": {
+ "ollama": "OLLAMA_BASE_URL",
+ "watsonx": "WATSONX_URL"
+ },
+ "project_id": {
+ "watsonx": "WATSONX_PROJECT_ID"
+ }
+ }
+
+ for field, mapping in field_mappings.items():
+ if field in template:
+ target_value = mapping.get(provider)
+ if target_value:
+ template[field]["value"] = target_value
+ template[field]["load_from_db"] = True
+ else:
+ template[field]["value"] = ""
+ template[field]["load_from_db"] = False
+ updated = True
+
+ return updated
+
+ async def _enable_model_in_langflow(self, provider_name: str, model_value: str):
+ """Ensure the specified model is enabled in Langflow."""
try:
enable_payload = [{
"provider": provider_name,
@@ -1305,64 +1097,35 @@ class FlowsService:
"enabled": True
}]
- enable_response = await clients.langflow_request(
+ response = await clients.langflow_request(
"POST", "/api/v1/models/enabled_models", json=enable_payload
)
- if enable_response.status_code == 200:
+ if response.status_code == 200:
logger.info(f"Successfully enabled model {model_value} for provider {provider_name}")
else:
logger.warning(
- f"Failed to enable model: HTTP {enable_response.status_code} - {enable_response.text}"
+ f"Failed to enable model: HTTP {response.status_code} - {response.text}"
)
except Exception as e:
logger.error(f"Error enabling model {model_value}: {str(e)}")
-
- # Update provider field and call custom_component/update endpoint
- if "model" in template:
- if "options" not in template["model"]:
- return False
-
- model = [template["model"]["options"][0]]
-
- template = await self._update_component_langflow(template, model_value)
-
- component_node["data"]["node"]["template"] = template
-
- model = [item for item in template["model"]["options"] if item["provider"] == provider_name and item["name"] == model_value]
-
- template = await self._update_component_langflow(template, model_value)
-
- template["model"]["value"] = model
-
- component_node["data"]["node"]["template"] = template
-
- if len(model) == 0:
- logger.warning(f"Model {model_value} not found for provider {provider_name}")
- return False
- updated = True
-
-
- if "api_base" in template:
- if provider == "ollama":
- template["api_base"]["value"] = "OLLAMA_BASE_URL"
- template["api_base"]["load_from_db"] = True
- elif provider == "watsonx":
- # Watson uses "url" field
- template["api_base"]["value"] = "WATSONX_URL"
- template["api_base"]["load_from_db"] = True
- else:
- template["api_base"]["value"] = ""
- template["api_base"]["load_from_db"] = False
- updated = True
-
- if "project_id" in template:
- if provider == "watsonx":
- template["project_id"]["value"] = "WATSONX_PROJECT_ID"
- template["project_id"]["load_from_db"] = True
- else:
- template["project_id"]["value"] = ""
- template["project_id"]["load_from_db"] = False
- updated = True
- return updated
+ def _get_provider_component_ids(self, provider: str):
+ """Helper to get component display names for various providers."""
+ from config.settings import (
+ OPENAI_EMBEDDING_COMPONENT_DISPLAY_NAME,
+ OPENAI_LLM_COMPONENT_DISPLAY_NAME,
+ WATSONX_EMBEDDING_COMPONENT_DISPLAY_NAME,
+ WATSONX_LLM_COMPONENT_DISPLAY_NAME,
+ OLLAMA_EMBEDDING_COMPONENT_DISPLAY_NAME,
+ OLLAMA_LLM_COMPONENT_DISPLAY_NAME,
+ )
+ if provider == "openai":
+ return (OPENAI_EMBEDDING_COMPONENT_DISPLAY_NAME, OPENAI_LLM_COMPONENT_DISPLAY_NAME)
+ elif provider == "watsonx":
+ return (WATSONX_EMBEDDING_COMPONENT_DISPLAY_NAME, WATSONX_LLM_COMPONENT_DISPLAY_NAME)
+ elif provider == "ollama":
+ return (OLLAMA_EMBEDDING_COMPONENT_DISPLAY_NAME, OLLAMA_LLM_COMPONENT_DISPLAY_NAME)
+ elif provider == "anthropic":
+ return (None, "Anthropic")
+ return (None, None)
diff --git a/src/utils/langflow_headers.py b/src/utils/langflow_headers.py
index bc8dac48f..dc21d619a 100644
--- a/src/utils/langflow_headers.py
+++ b/src/utils/langflow_headers.py
@@ -21,7 +21,7 @@ def add_provider_credentials_to_headers(headers: Dict[str, str], config) -> None
# Add WatsonX credentials
if config.providers.watsonx.api_key:
- headers["X-LANGFLOW-GLOBAL-VAR-WATSONX_API_KEY"] = str(config.providers.watsonx.api_key)
+ headers["X-LANGFLOW-GLOBAL-VAR-WATSONX_APIKEY"] = str(config.providers.watsonx.api_key)
if config.providers.watsonx.project_id:
headers["X-LANGFLOW-GLOBAL-VAR-WATSONX_PROJECT_ID"] = str(config.providers.watsonx.project_id)
@@ -53,7 +53,7 @@ def build_mcp_global_vars_from_config(config) -> Dict[str, str]:
# Add WatsonX credentials
if config.providers.watsonx.api_key:
- global_vars["WATSONX_API_KEY"] = config.providers.watsonx.api_key
+ global_vars["WATSONX_APIKEY"] = config.providers.watsonx.api_key
if config.providers.watsonx.project_id:
global_vars["WATSONX_PROJECT_ID"] = config.providers.watsonx.project_id