C3 AI Documentation Home

Changelog for genAiSearch

Table of Contents

  1. Changes to Release Process
  2. Upgrades
  3. New Features and Improvements
  4. Breaking Changes
  5. Other

Changes to Release Process

  • genAiBase and genAiSearch are now delivered as part of platform.
  • GenAI version inherited from platform release.

Applications that depend on GenAI

  • Must declare an explicit dependency on genAiSearch.
  • Teams can depend on version 8.9 of genAiSearch and will automatically use the version released as part of the platform.
  • Pin to minor version, not patch version.

New vanilla deploymnets

  • Create new app package when starting app in Studio
  • Declare 8.9 genAiSearch as dependency package

Upgrades

  • Upgrade Path: v6.1 on 8.8 → v8.9
  • Upgrade Script: Replace <env-id>, <app-id>, and <server-version> in below script and run as EnvAdmin
JavaScript
Env.forId('<env-id>').upgradeApp(Env.forId('<env-id>').app('<app-id>'), {
  serverVersion: '<server-version>',
  rootPkgVersion: '8.9.0',
});

New Features and Improvements

Agentic process automation

Agentic workflows enable repeatable task automation with standardized, reliable execution.

Documentation:

Canvas

The Canvas Agent combines the dynamic agent with full document-editing capabilities. Outputs appear directly in a rich text editor where users can continue editing or request changes from the agent.

  • Automatically deployed as part of quick start.
  • See Canvas Agent for configuration details

Deep research

The Deep Research Agent produces structured, comprehensive results for complex research-oriented questions.

  • Deploy using Genai.QuickStart.deployCoreAgents(['dynamic_canvas_deep_research'])
  • See Deep Research Agent for more details.

Files as inputs

Attach one or more files to a query to provide additional context for the agent.

  • Supports documents, images, and other common file types
  • See File Inputs for supported formats across LLMs

Improvements to metadata tagging

Improved creation, management, filtering of metadata tags and categories on Source Files. For more information, see Manual Metadata Tagging documentation.

Other

  • Integration with Studio for unified agent deployment and management.
  • Improved Mew3 performance by 40% through parallelized LLM calls and enhancements to table processing.
  • Expanded Genai.SourceFile.Chunker.Mew3.Config options by exposing more options for logging and LLM configuration.
  • Updated Python runtimes to 3.12 for improved ecosystem support.
  • Built prompts UI page for editing and add prompts.
  • Enhanced verbose mode to improve transparency, showing the actual code used by the agent to produce each AI summary result (documentation).
  • Added support for drafting, editing, and sending emails through the email agent.
  • Data Fusion integration: ingested data is automatically available in the default data model for structured retrieval queries

Breaking Changes

Feature removals and replacements

  • Self-service DI integration → Data Fusion in Studio
  • Old Chat UX (view mode chat) → Feed UX
  • User management page UI → Studio User management
  • Alerts → no longer supported

Deprecated features and replacements (planned for removal in next release)

  • Genai.Agent.QueryOrchestrator → GenaiCore.Agent
  • Genai.UnstructuredQuery.Engine → GenaiCore.Agent
  • Genai.Agent.Persistable and Genai.Agent.Dyanmic.Persistable →GenaiCore.Agent
  • All legacy tools (structured db agent, MSS, C3 method) → GenaiCore.Agent tools
  • Genai.Agent.Tool.Util.Attributor → attribution no longer supported
  • Genai.Agent.Tool.Util.Corroborator → no longer supported
  • Genai.SourceFile.Chunker.MultimodalPdf → Mew3
  • Genai.Retriever.Elser → Genai.Retriever.PgVector
  • Query & document translation → natively supported by dynamic agent and multi-lingual e5 embedder

Other breaking changes

  • Mew3 Configuration Schema: Breaking changes to the LLM client configuration require migration for deployments that use non-default LLM settings
  • Update Genai.SourceFile.Chunker.Mew3.Config#llmClient
  • See Multimodal Parsing for details.

Other

  • LLM keys should be configured at cluster level by administrator instead of per application for improved key management.
  • Mew3 does not support parallel processing on a single node. Scale horizontally by adding nodes to improve processing speed (provider rate limits may become bottleneck).
Was this page helpful?