Hi all,
As many of you know from my years working with and architecting Alfresco solutions, one of the biggest challenges our ecosystem faces today is the AI revolution. Everyone wants to build RAG (Retrieval Augmented Generation) pipelines, but securely feeding Enterprise Content and Workflow data into Vector Databases without bypassing Alfresco's strict security model is a massive headache.
To bridge this gap, I’m excited to share that we have released the first version of the Alfresco Repository Connector for OpenCrawling, a brand new, open source Data Federation engine I’ve been architecting from the ground up using Java 25, Spring Boot 4, and Virtual Threads.
Zero-Trust AI: It extracts documents (and their rich metadata) from ACS and strictly maps the Alfresco ACLs into the vector database. When an LLM answers a user prompt, it only retrieves context the user is actually authorized to see.
Workflow narrativization: Beyond static documents, we’ve extended the Open Ingestion Standard (OIS) to support BPMN 2.0. We are going to extract historic instances from APS/Activiti and flattening them into semantic, human-readable text chunks so AI agents can answer questions like:
- Why was invoice number 123456 rejected on May 25, 2020?
- Why wasn't the package delivered to the recipient for delivery under order number 987654321 dated April 14, 1985?
Cloud-Native Scale: Built to handle modern deployments, playing nicely with the latest stacks (like ACS/APS on OpenShift and Keycloak integrations).
We need your expertise: while the foundational connector is live, Alfresco environments are notoriously complex and highly customized. I am reaching out to my fellow Alfresco developers, architects, and power users to help us battle-test and refine this connector.
We are specifically looking for feedback and contributions on:
- Throughput & API Choice: Testing extraction speeds on massive document repositories. We'd love feedback on optimizing the CMIS vs. native REST API interactions using our Virtual Thread execution model.
- Complex Security Models: How well does the OIS ACL mapping hold up against highly customized Alfresco permission models and custom groups (especially with recent Keycloak setups)?
- Custom Metadata: Ensuring the connector elegantly handles deeply nested, custom aspect properties during the semantic embedding phase.
You can check out the code and the documentation here:
GitHub Repository: https://github.com/opencrawling/opencrawling/tree/main/oc-alfresco-repository-connector
Project Site: https://opencrawling.org
Whether you want to spin it up against a test environment, submit a PR, or just discuss the architecture of Enterprise AI, your feedback would be invaluable. Let's make Alfresco a first-class citizen in the LLM era!
Cheers, PG
