DCH 3.1
What's New
Version 3.1 of the DCH stack includes the following new features and improvements.
Agent Prompt Suggestion
Agents now include a built-in prompt generation system that automatically constructs a structured, operational brief for the language model.
This prompt defines the agent’s objective, grounds it in the connected graph data (memories and target entities), and specifies a clear method and expected output format. Rather than a generic instruction, it acts as a context-aware execution blueprint.
Prompts are dynamically generated based on the agent’s connected memories, ensuring that every analysis is aligned with the actual data model, relationships, and available context.
Linked Data API v2
Linked Data API v2 is now available. All endpoints have been rebuilt from the ground up with optimized Cypher queries, delivering improved performance and stability.
Endpoints superseded by v2 are marked as deprecated and will remain available for two additional versions before removal.
Affected Endpoints
| v1 Endpoint | Replaced by |
|---|---|
/api/graph/agent-context | /api/v2/graph/agent-context |
/api/graph/cypher/count | Functions |
/api/graph/cypher/nodes | Functions |
/api/graph/query/edges | /api/v2/graph/relationships |
/api/graph/query/edge-labels | — |
/api/graph/query/sub-graph | /api/v2/graph/related |
/api/graph/query/shortest-path | /api/v2/graph/shortest-path |
/api/graph/query/schema-visualization | — |
/api/graph/node-labels | /api/v2/graph/node-labels |
/api/graph/node-ids | /api/v2/graph/node-ids |
/api/graph/relationships | /api/v2/graph/relationship-definitions |
/api/graph/navigation/nodes | /api/v2/graph |
/api/graph/query/related-nodes | /api/v2/graph/related |
/api/graph/query/related-nodes/count-per-label | /api/v2/graph/related/count-per-label |
/api/graph/search/advanced-search | /api/v2/graph/search |
/api/graph/search/advanced | /api/v2/graph/search |
/api/graph/query/views/{idOrAlias} | /api/v2/views/{idOrAlias}/execute |
Neo4j Processing Improvements
Data processing from the staging layer to Neo4j has been significantly optimized for better performance and reliability:
- Reduced downtime: Each entity's downtime is now limited to its own processing duration rather than the entire job. Entities are cleared immediately before processing, ensuring that already-completed entities and relations remain intact if a job fails.
- Smarter prioritization: Entities with more relationships are processed first, unblocking dependent relation work items sooner.
- Parallel processing: Independent entities and relationships are now processed in parallel, reducing overall processing time.
Removed feature: Support for S3 and ABS processing modes has been discontinued. These modes previously used Amazon S3 and Azure ABS to buffer data during Neo4j processing. Due to performance improvements, native mode is now the most efficient and reliable option and is the only supported mode going forward.
Intake Job Parallelization
Intake jobs now run in parallel, enabling faster processing of large volumes of independent data. Parallelization is determined by the selected entities in each job. Additionally, the join logic has been optimized to speed up the import process.
Updated Docling in File Extraction Worker
The File Extraction Worker has been upgraded to the latest Docling version, delivering improved stability and faster processing when handling files through the File Intake Agent.