C3 AI Documentation Home

Changelog for genAiSearch

Table of contents

  1. New features
  2. Improvements
  3. Breaking changes
  4. Other

New features

Document generation

Workflows now support document authoring as a step in business process pipelines. Use the document generation node to produce formatted documents from structured data and AI-generated content. See Use Workflows for Document Generation.

Improvements to workflows

  • Added agent skills to improve output accuracy.
  • Added the ability to select a specific entity instance to pass to a workflow during testing.
  • Added a delay node to pause execution during long-running or asynchronous workflow steps.

Agent skills

Skills are packaged capabilities that agents load on demand to complete specialized tasks. Each skill bundles instructions, reference material, and executable scripts. Agents discover available skills automatically and run them in an isolated workspace. See DynamicAgent-07-Skills tutorial.

Agent long-term memory

Agents now persist and recall information across sessions. The platform creates memories automatically from conversations and surfaces them in future queries. Both user-specific and global memories are supported. See Agent Long-Term Memory.

Improvements

  • Canvas and Workflows are now accessible directly within Studio.
  • Added support for .xls and .ppt file formats in document ingestion.
  • Added structured output support for agents, which returns type-safe responses from LLM calls.
  • Added a stop button to cancel in-progress queries.
  • Improved the unstructured data ingestion (UDI) backend. Each Genai.SourceCollection can now reference a distinct Genai.UnstructuredPipeline, which supports different parsing, metadata tagging, and retrieval configurations per collection. See Configuring Unstructured Pipelines.
  • Improved Mew3 multimodal document processing:
    • Added VLM (vision language model) layout parsing.
    • Added hierarchy detection to extract structured layout as passage metadata.
  • Added data model graph configuration in the agent workbench.

Breaking changes

Feature removals and replacements

The following features were deprecated in 8.9 and are removed in 8.10. The Type definitions remain but are no longer used.

  • Genai.Agent.QueryOrchestrator → GenaiCore.Agent
  • Genai.Agent.Persistable and Genai.Agent.Dynamic.Persistable → GenaiCore.Agent
  • All legacy tools (structured DB agent, MSS, C3 method) → Dynamic agent tools
  • Genai.UnstructuredQuery.Engine → Dynamic agent
  • Genai.Agent.Tool.Util.Attributor and Genai.Agent.Tool.Util.Corroborator → no longer supported
  • Genai.SourceFile.Chunker.MultimodalPdf → Mew3
  • Query and document translation → native translation support through agents
  • Legacy Agents page → Agents in Studio
  • Legacy LLM configuration page → LLMs configured at the cluster level by Ops; LLM configuration page in Studio
  • Projects → no longer supported
  • Genai.ChatBot → Genai.Agent.Dynamic.Core#run or GenaiCore.Agent.Deployment#run
  • Genai.Agent.Dynamic.Config → Genai.Agent.Dynamic.InitializationSpec

Other

  • All unstructured processing code moved from genAiBase to the unstructuredDI package. Continue to depend on genAiBase.
  • The default LLM client is now default-completions, set by Ops at the cluster level. The default embedder is now the cloud embedder, also set by Ops at the cluster level.
  • Create agents through the Agents page in Studio rather than in Jupyter notebooks.
Was this page helpful?