RAG

RAG (Retrieval Augmented Generation) enables Devic assistants to access documentation and use it in real time during conversations.
Thanks to this capability, assistants can provide more accurate, grounded, and up-to-date responses, without relying solely on the model’s pretrained knowledge.

---

​

How It Works

Devic’s RAG system is fully integrated with assistants and does not require complex configuration.
When documents are added to an assistant’s knowledge base:

1. The platform automatically analyzes the content.
2. It splits the text into fragments (“chunks”) to improve contextual retrieval.
3. It generates summaries that are injected into the model’s context.
4. During the conversation, the assistant decides when to consult these documents and uses the most relevant fragments to build its response.

This approach provides the assistant with extended memory, allowing it to contextualize answers using user-uploaded materials—such as manuals, reports, or internal guides.

---

​

Configuring RAG in an Assistant

To add knowledge to an assistant:

1. Open the assistant from Devic’s sidebar.
2. Navigate to the Knowledge section.
3. Click Add document and select the files you want to upload

Supported formats:

• .pdf
• .docx
• .txt
• .csv

Each document is processed automatically and displayed in the interface with its name and file type.

---

​

Document Details and Metadata

You can access each document’s details by clicking on it.
The side panel will display its key information and metadata: The Summary field is used internally to inject contextual information into the assistant’s prompt.
You can edit it to highlight the most relevant data for your use case.

---

​

Document Preview

From the same detail panel, you can preview the full content of the document directly in Devic—no download required. This feature is useful for reviewing information before starting a conversation or refining document summaries.

---

​

Best Practices

• Ensure the Search Knowledge tool is enabled if the assistant needs to use RAG.
• Upload only documents that are relevant to the assistant’s scope.
• Review automatically generated summaries and edit them if needed.
• Use descriptive file names.
• Keep documents updated to ensure coherent responses.
• Avoid uploading excessively large or unstructured files.

---

​

Use Cases

---

​

Practical Example

In this example, the assistant Alexandria has several technical documents loaded and can answer questions related to user manuals:

1. The user uploads 4011690_Manual Operador_e.pdf.
2. Devic automatically generates a summary with key points.
3. During the conversation, Alexandria uses RAG to retrieve relevant fragments from the manual and provide precise instructions.

---

​

Summary

---

​

Next Steps