What's New inKaryam

This release introduces File Upload Support in Agent Chat, enabling users to attach documents, data files, and images for AI-assisted analysis and responses.

What’s New

File Upload Support

Attach up to 5 files in a single chat message AI agents can read uploaded files and use their contents as context for responses Supports both documents and images

Supported File Types

Documents: .pdf, .md, .txt
Data Files: .json, .log
Images: .png, .jpg, .jpeg

Agent Configuration

To enable file uploads for an agent:

Enable the Enable File Upload option in the Agent configuration Add the File Read skill to the agent

File uploads are available only when both Enable File Upload is enabled and the File Read skill is configured for the agent.

Provide richer context to AI conversations by allowing agents to analyze uploaded files directly within Agent Chat.

This release introduces a major upgrade to Karyam’s Knowledge Platform, bringing production-ready Retrieval-Augmented Generation (RAG), multiple vector database providers, automated embedding pipelines, and enhanced document ingestion. These improvements make AI agents more accurate, scalable, and easier to deploy in production.

Release Highlights

• Introduced Knowledge Bases for organizing documents, datasets, and business knowledge into reusable AI-ready repositories.
• Added a production-ready Retrieval-Augmented Generation (RAG) pipeline with automatic document parsing, chunking, embedding generation, indexing, and semantic retrieval.
• Added support for multiple Vector Database Providers, allowing organizations to choose the storage backend that best fits their infrastructure.
• Introduced configurable Embedding Models that can be assigned independently to each Knowledge Base.

Vector Database Support

• Added native support for Qdrant.
• Added support for PostgreSQL with pgvector for teams that prefer using their existing PostgreSQL infrastructure.
• Unified vector storage APIs across supported providers for a consistent indexing and retrieval experience.
• Added automatic storage validation and initialization during Knowledge Base creation.
• Improved embedding dimension validation to prevent model and vector database incompatibilities.

Knowledge Base Improvements

• Added support for uploading documents directly into Knowledge Bases.
• Introduced automatic document chunking with optimized chunk sizing for improved retrieval quality.
• Added metadata indexing, including source files, chunk identifiers, and document references.
• Improved document synchronization and incremental updates without rebuilding the entire Knowledge Base.
• Added automatic re-indexing whenever knowledge sources are updated.

RAG Pipeline Enhancements

• Introduced configurable retrieval strategies for higher-quality context selection.
• Improved semantic search accuracy through optimized embedding retrieval.
• Added source attribution so AI responses can reference the originating document.
• Optimized embedding generation for faster ingestion of large document collections.
• Reduced retrieval latency through improved vector search workflows.

Document Processing

• Improved document parsing across supported file formats.
• Added intelligent preprocessing before chunk generation.
• Enhanced handling of large documents through streaming ingestion.
• Improved metadata extraction for more relevant semantic search results.

Developer Experience

• Added configurable embedding providers within workspace settings.
• Improved logging and visibility for document ingestion and embedding jobs.
• Enhanced vector database configuration with provider-specific validation.
• Simplified switching between vector database providers without changing application logic.

Impact

This release provides a solid foundation for building enterprise AI applications by delivering:

• Faster and more reliable document ingestion.
• Higher-quality semantic search and retrieval.
• Flexible deployment using multiple vector database providers.
• Production-ready infrastructure for AI agents, assistants, and enterprise knowledge systems.
• Improved scalability and consistency across Knowledge Bases.

Recommended Configuration

For the best retrieval quality, we recommend:

• Use a high-quality embedding model appropriate for your domain.
• Organize documents into focused Knowledge Bases.
• Keep source documents up to date to ensure fresh retrieval results.
• Choose the vector database that best matches your deployment:
  • Qdrant for dedicated high-performance vector search.
  • PostgreSQL + pgvector for unified relational and vector storage.

---

This release marks an important milestone in Karyam’s AI platform, enabling developers to build more intelligent, context-aware, and production-ready AI applications powered by enterprise knowledge.