The qKnow Agent Construction Platform Open Source Edition v2.4.3 focuses on upgrades regarding unstructured knowledge ingestion, knowledge asset export, and customizable document parsing. Key updates include:
- New Import Capabilities: Support for importing JSON/JSONL files into Knowledge Bases and Knowledge Graphs.
- Data Export: One-click export of Q&A data from Knowledge Bases to JSON/JSONL formats.
- Customization: Opened capabilities for customizing file parsing models and prompts.
These updates further refine the processing pipeline for unstructured documents, covering ingestion, parsing, and reuse.
From "Uploading Documents" to Enabling Data Ingestion, Processing, and Reuse
When enterprises build knowledge bases and agents, raw knowledge isn't limited to standard documents like Word or PDF files. Real-world projects often involve document parsing results processed by other systems, semi-structured data, and pre-organized Q&A datasets.
This means the enterprise knowledge processing workflow is not simply:
Upload File → Parse → Build Knowledge Base
It is closer to:
External Data Preparation → Data Import → Document Parsing → Knowledge Generation → Data Export → Secondary Processing & Reuse
When a platform's data entry/exit methods and parsing logic are too rigid, three issues often arise as scenarios increase:
- Lack of Import Entry: Parsed or organized JSON/JSONL data lacks a standard way to enter the platform.
- Difficulty Exporting: Q&A knowledge generated in the knowledge base cannot be easily exported for external processing or dataset accumulation.
- Rigid Parsing: Different document types require different parsing logic, but fixed models and prompts make it hard for users to adjust to specific business scenarios.
Therefore, the focus of qKnow Open Source v2.4.3 is not just adding file formats, but completing the full loop:
Data enters platform → Parsed per business needs → Knowledge formed → Results output for reuse
1. New JSON/JSONL Import for Knowledge Bases & Graphs
Previously, qKnow relied primarily on standard file uploads. However, in enterprise data processing, some unstructured documents may have already been parsed or converted into more structured formats like JSON or JSONL by other systems. If the platform forced users to start from raw files again, it would lead to redundant work.
This version adds JSON/JSONL import capabilities to provide a standard entry point for this data.
Direct Upload Support
The update supports uploading JSON and JSONL files. This means data that has already been organized doesn't need to be reverted to raw documents; it can be imported directly based on its actual form.
The workflow now supports:
Raw Document → External Parsing/Processing → JSON/JSONL → qKnow Knowledge Base/Graph
This is particularly suitable for enterprises with existing document processing pipelines or those needing to ingest data generated by other systems.
Distinguishing Data Sources
The update distinguishes between unstructured and semi-structured data to adapt to different ingestion scenarios. While JSON/JSONL has structure, the content may originate from unstructured document parsing or be pre-organized semi-structured knowledge. This allows different forms of knowledge data to enter the processing flow.
Bridging the Gap for Re-ingestion
A key application is re-ingesting data that has already been parsed. For example, if an enterprise uses external tools to parse files and saves the results as JSON, qKnow can now import this directly, avoiding the cycle of converting results back to files just to re-parse them.
Note: The JSON/JSONL import solves the ingestion problem. How this data flows into knowledge parsing and construction depends on the specific data format and the enterprise's configuration.
2. Export Q&A Data to JSON/JSONL for Reuse
A knowledge platform must address not only how data enters but also how it leaves.
In enterprise knowledge base construction, parsing a document often generates a large amount of Q&A knowledge. This data is valuable for:
- External system processing.
- Dataset organization.
- Model evaluation.
- Data annotation.
- Cross-system reuse.
If parsing results are trapped inside the platform, knowledge assets cannot flow freely.
One-Click Export
qKnow Open Source v2.4.3 adds the ability to export Q&A data. After parsing documents and building the knowledge base, users can export the Q&A data into standard JSON or JSONL files.
The workflow becomes:
Upload → Parse → Generate Q&A → Build KB → Export JSON/JSONL → External Processing
Standardized Output for Dataset Accumulation
For teams building enterprise agents long-term, Q&A data is a sustainable asset. It can be used for verification, test set construction, training data preparation, or manual review. Standardized export formats facilitate batch processing better than manual copying.
From "KB Results" to "Tradable Assets"
This shifts the paradigm from:
File → Parse → Internal Use
To:
File → Parse → Knowledge → Internal Use → Export → External Reuse
This increases the flexibility of document-based knowledge assets across different tools. However, exported data may still require quality checks or format adjustments depending on the target use case (e.g., model training).
3. Customizable Parsing Models and Prompts
Enterprise documents vary significantly—policy documents differ from technical manuals, contracts, or maintenance logs.
- Technical docs might focus on: Equipment, parameters, steps, exceptions.
- Policy docs might focus on: Scope, rules, responsibilities, execution requirements.
A fixed parsing logic often fails to accommodate these differences. Previously, qKnow used fixed system logic for models and prompts.
Open Model Selection
Users can now adjust the model used for document parsing based on business needs. Parsing is no longer solely dependent on a preset platform model but can leverage the enterprise's own model resources.
The logic flows as:
Identify Doc Type → Select Parsing Model → Configure Prompt → Execute Parsing → Review Results
Configurable Prompts
The update also allows for custom parsing prompts. Prompts directly influence:
- What content to extract.
- How to understand the document.
- How to organize the generated results.
- Which business information to prioritize.
This allows the same model to adjust its extraction targets based on the scenario (e.g., focusing on main content vs. specific business entities).
Adapting to Scenarios
This shifts the workflow from a rigid process to a flexible one:
Upload → Select Model → Configure Prompt → Parse → Use Results
This is ideal for enterprises with diverse document sources. However, customization solves the adjustability of logic; actual results still depend on document quality, model capability, and prompt design.
4. A More Open Knowledge Processing Pipeline
Combining these three features, qKnow v2.4.3 establishes a more complete knowledge data pipeline.
Previous Flow:
Standard File Upload → Fixed Parsing → Knowledge Base → Internal Use
New Flow:
Raw File / JSON / JSONL → Import → Custom Model/Prompt Parsing → KB Q&A Data → JSON/JSONL Export → External Reuse
- JSON/JSONL Import: Solves how different data forms enter.
- Custom Parsing: Solves how knowledge is processed according to business logic.
- JSON/JSONL Export: Solves how processed knowledge flows out for reuse.
Version Value Summary
This release makes unstructured knowledge handling more flexible from ingestion to reuse:
- Open Ingestion: JSON/JSONL import reduces redundant conversion for pre-processed data.
- Asset Circulation: Export capabilities allow knowledge to be used for external processing and dataset accumulation.
- Controllable Logic: Custom models and prompts allow parsing to adapt to specific document types and business goals.
Overall, qKnow Open Source v2.4.3 connects the "Import — Parse — Export" loop, giving enterprises more autonomy in building and reusing non-structured knowledge.
Top comments (0)