Changes In Release 26.10 For Agent Factory
All content in this Agent Factory User’s Guide has been updated to reflect the functionality and user interface changes introduced in release 26.10.
New Features and Updates
The following sections list the new features and updates in Agent Factory release 26.10.
Main Features
| Feature | Description |
|---|---|
| Native Observability Dashboards | Agent Factory includes built-in dashboards for metrics, sessions, traces, and spans. Run an agent, review its metrics and sessions, and drill down to trace and span details. |
| OpenRouter Node in Agent Builder | Use a saved OpenRouter configuration as the LLM source for an Agent node. Optionally set a system message or override the model ID for the node. |
| Vector Search Node in Agent Builder | Search content prepared by an ingestion, then pass results to downstream nodes or expose vector search as a tool to an Agent Builder Agent. The default result limit is 5. |
| Deep NL2SQL | Create an agent for selected database tables and views, optionally add a published Business Knowledge Graph for business context, ask questions and resolve clarifications, and review generated SQL, results, and agent activity. |
| Memory Workbench | Review clarification answers saved as memories from Deep NL2SQL conversations, and delete one memory or all memories. |
| Agent Memory Package Upgrade, Enable Agent Memory for a Workflow | Agent Factory uses Oracle Agent Memory 26.6 to extract durable memories in the background after successful chats. Workflow builders can configure custom extraction instructions, and authorized users can edit or delete memory cards in Inspect Memory. Existing compatible Agent Memory data is preserved during the package-managed upgrade. |
| DB Agent Engine | Agent Builder can run compatible workflows in an authorized Oracle AI Database PDB. Administrators and authorized editors configure an execution database and compatible OCI Generative AI model, validate readiness, and then enable DB Agent Engine for a workflow. Map, Parallel Map, and Parallel Flow can run compatible saved custom flows sequentially or concurrently. |
| Agent Factory Python SDK | Python applications can authenticate with a username and password or configured SSO, run authorized Custom Workflows and pre-built agents, and manage supported data sources and model connections. Agent Factory checks SDK release compatibility and records correlation information for diagnostics. |
| REST API Integration Options | External applications can use REST APIs to manage supported model and data source configurations, list users, and run supported agents. Integrations can authenticate with an agent-specific API key, local HTTPS Basic credentials, or a deployment-scoped Service Account API key, depending on the operation. |
| MCP Server Authentication Enhancements | MCP servers support Basic authentication, shared or per-user OAuth, durable authorization pause and resume, and custom downstream request headers. |
New Data Source, Knowledge, and RAG Agent Updates
| Feature | Description |
|---|---|
| AI Knowledge Store | Associate database sources with a store for semantic metadata, annotations, and Business Knowledge Graph content. |
| Knowledge Workbench | Create Business Knowledge Graphs, add business context, evaluate them with natural-language and SQL examples, and publish them for agent use. |
| Annotation Workbench | Add business descriptions and annotations to database objects to provide semantic context for supported AI features. |
| RAG Agent terminology update | The user interface and guide use the RAG Agent name for this pre-built agent. |
| Image Support in RAG Agents | Use standalone JPG, JPEG, and PNG images from supported data sources as context in RAG Agent responses with compatible OCI Generative AI models. Ask image-related questions and open cited images from the response’s Sources list. |
| Reusable Ingestion Service | Create reusable ingestions from supported data sources, monitor processing and versions, and refresh content without interrupting access to the last successful version. |
| Confluence Data Sources | RAG Agents can ingest pages from Confluence Cloud or Confluence Data Center by using OAuth authentication and a configurable root-page crawl scope. |
| Queued Ingestion Status | The INGESTION_QUEUED data source status indicates that a source is queued for ingestion, but processing has not started. |