RAG Ingestion

In Cencori, Vector Memory (RAG) is distinct from Adaptive Memory (Observational).

• Adaptive Memory: "Who the user is" (Short/Medium Context).
• Vector Memory: "What the user knows" (Long Context / Documents).

You can use Workflows to build automated ingestion pipelines.

The Strategy

Instead of manually uploading files, you can listen for file.uploaded events and automatically process them.

1. Trigger: file.uploaded (or a custom webhook).
2. Process: Extract text (PDF/DOCX to Markdown).
3. Chunk: Split into semantic chunks.
4. Embed: Generate vectors (Cencori handles this).
5. Persist: Save to cencori.memory.vector.

Implementation

1. The Trigger

cencori.on('file.uploaded', async (event) => {
  await cencori.workflows.trigger('rag-ingestion', event);
});

2. The Extraction Step

Use a tool (like unstructured or pdf-parse) to get clean text.

// Pseudocode for Workflow Step
const text = await tools.extractText(event.file.url);

3. The Injection Step

Push the text into Cencori's Vector Store.

await cencori.memory.vector.add({
  content: text,
  metadata: {
    source: event.file.name,
    user_id: event.user.id
  }
});

The Result

Now, any file the user uploads is instantly searchable by your agent.

// Retrieve context
const context = await cencori.memory.vector.search(userQuery);

This pipeline ensures your RAG knowledge base is always in sync with your files.