Skip to main content
Version: Current

Conversational Agent Integration

The IB-X Knowledge Ingestion Service integrates directly with Conversational Agents to provide grounded, context-aware, and semantically relevant AI responses.

Knowledge ingested through the Knowledge Ingestion Service becomes part of the semantic retrieval context associated with the owning Agent.

During conversation, the Conversational Agent retrieves semantically relevant knowledge from its configured ingestion sources and supplies the retrieved context to the AI model before response generation.

This architecture enables Retrieval-Augmented Generation (RAG) experiences within the IB-X platform.


Agent-Specific Knowledge Isolation

Knowledge ingestion in IB-X is Agent-specific.

Each Conversational Agent maintains its own isolated semantic knowledge space consisting of:

  • Ingestion sources
  • Semantic embeddings
  • Retrieval context
  • Ingestion runs
  • Specialized collections

Only the owning Agent can retrieve and use its ingested knowledge.

This isolation model helps improve:

  • Retrieval relevance
  • Knowledge ownership
  • Domain specialization
  • Security boundaries
  • Response quality

Retrieval Flow

During conversation, the Conversational Agent performs semantic retrieval against its associated knowledge space.

The retrieval workflow typically follows the sequence below:

  1. User submits a question or message
  2. Conversational Agent performs semantic similarity search
  3. Relevant embeddings are identified
  4. Related semantic chunks are retrieved
  5. Retrieved context is supplied to the AI model
  6. AI model generates a grounded response
  7. Response is returned to the user

This retrieval pipeline enables the Agent to generate responses using organization-specific knowledge instead of relying only on foundational model knowledge.


Semantic Retrieval

The Knowledge Ingestion Service stores semantic embeddings inside the configured Vector Store.

During retrieval, the Conversational Agent performs semantic similarity search to identify the most contextually relevant knowledge chunks.

Semantic retrieval allows the platform to retrieve information based on:

  • Meaning
  • Intent
  • Context
  • Conceptual similarity

instead of relying only on keyword matching.


Grounded Responses

The retrieved semantic context is supplied to the AI model before response generation.

This grounding process helps the Agent generate responses that are:

  • Context-aware
  • Organization-specific
  • Knowledge-driven
  • More accurate
  • Less hallucinated

Grounding significantly improves enterprise conversational reliability and retrieval quality.


Specialized Collection Usage

The Knowledge Ingestion platform supports specialized semantic collections that improve retrieval organization and contextual relevance.

Supported collections currently include:

CollectionPurpose
FAQsFrequently asked questions
ProductsProduct-specific information
API ReferencesTechnical and API documentation
Code SnippetsSource code and implementation examples

These collections help improve downstream semantic retrieval quality for targeted conversational scenarios.


Benefits

Using the Knowledge Ingestion Service with Conversational Agents provides several enterprise advantages.

Context-Aware Responses

Agents can retrieve and respond using organization-specific knowledge.


Reduced Hallucinations

Grounded retrieval reduces dependency on generalized foundational model knowledge.


Retrieval is based on contextual meaning instead of only keyword matching.


Domain Specialization

Each Agent can maintain its own specialized semantic knowledge space.


Enterprise Knowledge Grounding

Organizations can ground AI experiences using internal documentation, websites, files, videos, and other knowledge sources.


Supported Knowledge Sources

Conversational Agents can retrieve knowledge ingested from:

  • Website URLs
  • Uploaded files

The supported ingestion workflows are documented separately.



Notes

  • Knowledge retrieval is isolated to the owning Agent.
  • Semantic embeddings are stored in the configured Vector Store.
  • Relationship-based retrieval scenarios may use the configured Graph Database.
  • Ingestion infrastructure is configured globally from the root tenant.