GuidesAPI ReferenceChangelogOpen RoadmapStatusSign up
Guides

Create your RAG project

Eden AI makes it easy to build a custom RAG (Retrieval-Augmented Generation) chatbot powered by the best AI models on the market. You can ingest your own data, choose your preferred AI providers, and deploy an API-powered chatbot in minutes.

Suggest Edits

🚀 Getting Started

When creating a RAG project, you have two options:

✅ Quickstart

Start with Eden AI’s default settings, optimized for fast deployment. You don’t need to configure anything — just upload your data and go.

⚙️ Create from Scratch

Advanced users can manually select each component (vector DB, embeddings, LLM, etc.) for full control and customization.

🔧 Step-by-Step: Create from Scratch

1. Choose RAG Technique

  • Currently, the basic RAG technique is available: retrieve relevant chunks and pass them to the LLM to generate an answer.
  • In the near future, Eden AI will support advanced RAG techniques:
    • Self-RAG
    • Graph-RAG
    • Long-RAG
    • ... and more!

2. Select a Vector Database (Database Provider)

Store and search your documents efficiently using vector embeddings. You can choose between:

  • Eden AI DB (fully managed by Eden AI)

    • Qdrant ✅ Recommended
    • Supabase
    • (More to come: Pinecone, Weaviate, etc.)

  • Your own DB: Bring your own account and connect to external vector stores.

3. Embeddings Provider

Embeddings convert your text into vector representations that the database can search through. Choose from the top providers:

  • Cohere
  • Mistral
  • Jina
  • OpenAI
  • Google

4. LLM Model

This is the model that will generate the final response using retrieved documents. Eden AI supports all major LLM providers:

  • OpenAI
  • Anthropic
  • Google
  • Mistral
  • Meta
  • xAI
  • Deepseek
  • Amazon
  • ...and more

Choose the model that fits your use case in terms of performance, latency, and cost.

📘

Don't worry! You can always switch to another model when using your RAG chatbot!

5. Set Chunk Size

Chunk Size determines how your documents are split for indexing.

  • It’s the maximum number of characters in each segment (chunk).
  • A typical value is 1200, which balances context and performance.
  • Smaller sizes make searches faster and more relevant for short queries.
  • Larger sizes keep more context but may affect performance.

6. Configure Chunk Separators

Chunk Separators define where the document is split into chunks.

  • Examples: \n\n, \n, ., ", etc.
  • They allow you to split text cleanly at paragraph breaks, sentence ends, or custom tokens.
  • You can combine multiple separators for better results.

7. Optional: OCR Model

If your documents include images or PDFs, OCR (Optical Character Recognition) will extract text for indexing.

  • Available provider: Amazon, Microsoft
  • More providers coming soon.

8. Optional: Speech-to-Text Model

If your files include audio or video, use Speech-to-Text to transcribe them.

9. Create Your Project

Once everything is set, click "Create Project" — You will now be able to start indexing your content, and you’ll be ready to use your chatbot via API.

Updated 1 day ago