Using RAG: Give Agents Your Knowledge
RAG (Retrieval-Augmented Generation) is the technology that lets agents search your Library during conversations. Instead of just relying on their training data, agents can pull in relevant content from your files to give more accurate, context-aware responses.
How RAG works
When you upload a file to the Library, it goes through an indexing pipeline:
- Upload — Your file is stored securely in cloud storage.
- Processing — The content is extracted and split into chunks.
- Embedding — Each chunk is converted into a vector embedding using AI.
- Indexing — Embeddings are stored in a vector database (Qdrant) for fast similarity search.
During a conversation, when an agent determines it needs additional context, it searches your indexed files for the most relevant chunks and includes them in its response generation.
Scoping: Personal vs. Project
- Personal library: Only available to you in your conversations.
- Project library: Available to all project members during conversations within that project. Upload project-specific documentation, specs, or reference material here.
What works best with RAG
- Technical documentation, API references, and code examples
- Meeting notes, project specs, and requirements documents
- Research papers, articles, and reference material
- Style guides, brand guidelines, and templates
Tips for better results
- Use descriptive file names — agents see file metadata during search.
- Break large documents into focused files — a 200-page manual is harder to search than 20 focused chapters.
- Keep content current — re-upload or edit files when information changes; the index updates automatically.
- Check processing status — newly uploaded files show a processing indicator. Wait until indexing completes before expecting the agent to reference them.