References

FAQs & Troubleshooting

How does the platform provide transparency and inventory for deployed AI Agents?
The Expert Agent Studio showcases a complete inventory of all deployed Agents, workflows and Multi-agent Systems. This allows users to easily view and manage all active AI assets, providing necessary transparency for governance and oversight.

Does turning Hybrid Search off in my agent change the Knowledge Garden?
No. It only changes how this agent retrieves. The Knowledge Garden, its indexes and other agents using it are not affected, and nothing is re-indexed.

Why don't I see the Hybrid Search toggle?
The attached Knowledge Garden doesn't have Hybrid Search enabled. Ask its owner to enable it in the Knowledge Garden configuration.

What is the default for Hybrid Search?
On. When you attach a Knowledge Garden that supports Hybrid Search, the toggle is on by default.

Will existing agents behave differently after this release?
No. Existing agents and Knowledge Gardens keep their current retrieval behavior.

Does turning Hybrid Search off make retrieval faster?
It doesn't add latency. Retrieval stays within the platform's usual performance limits either way.

Hybrid Search Troubleshooting

Error / Issue	Probable cause	Resolution
Toggle not shown for a Knowledge Garden	Hybrid Search isn't enabled on that Knowledge Garden	Enable it in the KG configuration (EKG → Settings → Retrieval Strategy)
Exact IDs or codes aren't being found	Hybrid Search is off for that Knowledge Garden	Turn the toggle on, then test again
Answers include loosely related chunks with matching keywords	Keyword matches are adding noise for conceptual questions	Turn Hybrid Search off for that Knowledge Garden, or adjust Top-K and the re-ranker
Not sure which mode was used	—	Check the retrieval details in Traces

Limitations

Type	Limitation (with explanation)
Default / RAG / ReAct (Conversation Agent)	Copied output is in Markdown format – When copying responses from a Conversational Agent, the text is automatically copied in Markdown formatting.
Default Automation Agent	Table output parameters work best with GPT‑4o‑16K – For optimal accuracy and formatting when returning table data, use the GPT‑4o‑16K model in automation mode.
Default / RAG / ReAct	Language restriction for file processing – On‑demand document ingestion currently supports English‑language files only.
ReAct	Structured action inputs required – In the Experience configuration, both Action and Action Input must start with `{}` to define structured parameters; natural language inputs are not supported.
Default / RAG / ReAct	No direct file import from Google Drive / OneDrive – Files must be uploaded manually to the platform; direct linking from external cloud storage is not yet supported.
Default / RAG / ReAct	Maximum file size limits – Document files must be under 20 MB and images under 5 MB to be processed successfully.
Default / RAG / ReAct	Unsupported file security features – Encrypted, zipped, or password‑protected files cannot be processed.
Default / RAG / ReAct	Supported formats only – File ingestion supports PDF, PNG, JPEG, TIFF, DOCX formats.
Default / RAG / ReAct	Logprobs only on Azure GPT‑4o models – Confidence scoring is available only on Azure GPT‑4o models, limiting debugging and analysis capabilities for other models.
RAG / ReAct	Vision models max 20 images – Image‑enabled models process only up to 20 images per query, limiting multi‑image analysis capabilities.
ReAct	Some connectors are read-only – Certain connectors allow only read operations, preventing direct data updates from the platform (e.g., current ServiceNow and IMAP connectors support only fetching, not creating or updating records).
Default/RAG/ReAct	Table as output for automation agent is only available for gpt-4,4o models only
RAG / Prime	Hybrid Search at the agent level is a Beta feature – The toggle is available only for Knowledge Gardens that have Hybrid Search enabled in their configuration. If a Knowledge Garden doesn't have it, the agent can't use hybrid retrieval on it.
Default/RAG/ReAct/Prime	Azure GPT - 5 model does not support change in Temperature and Top P. The model itself doesn’t support these changes - https://community.openai.com/t/temperature-in-gpt-5-models/1337133

List of Agent Add-ons

Guardrails - Toxicity Detection

What it does: This add-on enables advanced content filters to identify and mitigate toxic, harmful, or inappropriate content. It actively screens both user inputs and agent-generated responses.
Consideration: Overly lenient settings may allow harmful or biased content; configure filters carefully to uphold safety and respect.

Conversation Starters

Provides an initial set of suggested questions or topics to help the user begin an interaction with the Digital Expert, guiding them toward relevant areas of inquiry and reducing the effort needed to formulate a first request.

Follow up questions

Suggests up to three relevant follow‑up questions after a user query to guide deeper engagement and exploration.

Example in Purple Fabric: After a user asks "Show me last quarter's sales performance," the agent might suggest:

"Would you like a breakdown by region?"
"Do you want to compare with the previous year?"
"Shall I highlight the top‑performing products?"

Traces

Lets you see the detailed steps the agent took to arrive at its answer, shown directly in the response.

This only controls whether the trace steps are visible to you. Purple Fabric always keeps a full, permanent record of every action for audit and compliance, even if traces are not shown.