Workflow Change Log 1.2

OpenAI is now used as the default primary model instead of AWS Bedrock

LLM call

Step - GraphDB MCP - Async

Step - Input Guardrails - Sync

Improvement

No

-

Guardrails and answers now handle queries more gracefully when no relevant context is retrieved.

Step - Output guardrails - Async

Parallel Post Final Answer - Main

Step - Final Answer - Sync

Improvement

No

-

An error thrown by triggered input guardrails was remedied

Main

Bug fix

No

-

References from the final answer were removed

Step - Final Answer - Sync

Improvement

No

-

Optional dataset context for output guardrails was added

Configuration

Step - Output guardrails - Async

Improvement

No

Optional:

Use the configuration property "datasetContext" if the output guardrail behavior needs to be adjusted for your dataset

Guardrails and answers for off-topic questions were improved

Step - Final Answer - Sync

Step - GraphDB MCP - Async

Step - Input Guardrails - Sync

Improvement

No

-

Answers now use the correct Markdown format

Step - Final Answer - Sync

Bug fix

No

-

The embeddingModelId parameter mandatory in the Configuration Json was defined

Configuration, Step - Vector Search - Async

Vector search tool

Improvement

No

Mandatory:

embeddingsModelId is now mandatory in the Configuration json

Vector search was added as an MCP tool

Configuration, Step - Vector Search - Async

Vector search tool

Parallel - Main

Step - GraphDB MCP - Async

Improvement

No

Mandatory

the vectorSearchTopK property is now mandatory in the Configuration Json, it controls number of documents returned by the vector search; default value is 5.

Behavior when vector search results were empty was remedied

Step - Vector Search - Async

Vector search tool

Bug fix

No

-

Sampling temperature was set to 0

Step - GraphDB MCP - Async

Bug fix

No

-

Grounding with vector search results improved follow-up questions

Parallel Post Final Answer - Main

Step - Followups generation - Async

Parallel - Main

Step - Vector Search - Async

Vector search tool

Parallel - Steps Messaging

Main, Parallel - Steps Batching

Step - GraphDB MCP - Async

Step - Final Answer - Sync

Improvement

No

-

Vector search async was improved by adding rephrasing of the question using the chat history

Parallel Post Final Answer - Main

Step - Followups generation - Async

Parallel - Main

Step - Vector Search - Async

Vector search tool

Parallel - Steps Messaging

Main, Parallel - Steps Batching

Step - GraphDB MCP - Async

Step - Final Answer - Sync

Improvement

No

Mandatory

Parallel - Steps Messaging has to be added to the N8N_WORKFLOW_OVERWRITE_PARALLEL_DATATABLE_ID env variable

LLM usage is now shown if a workflow fails

Parallel Post Final Answer - Main

Step - Followups generation - Async

Parallel - Main

Step - Vector Search - Async, Vector search tool, Parallel - Steps Messaging

Main, Parallel - Steps Batching

Step - GraphDB MCP - Async

Step - Final Answer - Sync

Bug fix

No

-

Renamed all workflows, credentials, nodes and variables containing PoolParty to Graph Modeling

Step - Input guardrails - Main

Step - Graph Modeling Concept Enricher - Async

Step - Short-term memory - Sync

Step - Short-term memory - Async

Step - Graph Modeling Concept Enricher - Sync

Step - Output guardrails - Async

Step - Input Guardrails - Sync

Step - GraphDB MCP - Async

Main, Fetch Concept Details

Error - Sync

Error - Async

Parallel - Main

Step - Graph Modeling Expansion - Async

Parallel Post Final Answer - Main

Improvement

No

Mandatory:

Step - Graph Modeling Concept Enricher - Async and Step - Graph Modeling Expansion - Async have to be added to the N8N_WORKFLOW_OVERWRITE_PARALLEL_DATATABLE_ID env variable