Skip to content
Regolo Logo
Regolo Logo

Agentic RAG with Regolo + Elysia: Build a “Second Brain” that reasons, not just retrieves

Curated byFrancesco Massa·Fresh as of

👉Try Regolo on Elysia

Standard RAG chatbots are dumb. They retrieve 5 chunks and summarize them, often missing the “big picture” or failing when you need aggregation (e.g., “count the invoices”).
To build a true “Second Brain,” you need an agent that decides how to search (keyword vs. vector), when to summarize, and when to show a table instead of text—without sending your private knowledge base to US clouds.

Deploy Elysia (Weaviate’s decision-tree framework) powered by Regolo’s 120B models in under 10 minutes. Get secure, local-first RAG that thinks before it answers.

Outcome

  • Reasoning over Retrieval: Elysia uses decision trees to pick tools (Query, Aggregate, Summarize). Regolo’s gpt-oss-120b or Llama-3.3 drives these decisions with high accuracy.
  • Zero-Data Retention: Your documents stay in your local Weaviate or private cloud. Regolo only sees the transient inference context, never training on your files.
  • Dynamic UI: The agent auto-selects the best view (Tables, Charts, Cards) based on your data, not just text walls.

Prerequisites (Fast)

  • Regolo API Key: From your dashboard.
  • Docker: To run Weaviate locally.
  • Python 3.12+: For the Elysia backend.​

Step-by-Step (Code Blocks)

1) Run Weaviate Locally (Self-Hosted Memory)

Don’t use a cloud DB if you want privacy. Run Weaviate on your machine.

# docker-compose.yml
services:
  weaviate:
    image: cr.weaviate.io/semitechnologies/weaviate:1.33.0
    ports:
    - "8082:8080"
    - "50051:50051"
    environment:
      QUERY_DEFAULTS_LIMIT: 25
      AUTHENTICATION_ANONYMOUS_ACCESS_ENABLED: 'true'
      PERSISTENCE_DATA_PATH: '/var/lib/weaviate'
      ENABLE_MODULES: 'text2vec-ollama,generative-ollama' # Modules for local vectors
    volumes:
    - weaviate_data:/var/lib/weaviateCode language: Dockerfile (dockerfile)

Command: docker-compose up -d
Expected output: Weaviate listening on http://localhost:8082

2) Install & Start Elysia

Elysia is the “brain” and UI. Install it via pip.

pip install elysia
elysia startCode language: Bash (bash)

Expected output: Elysia UI opens in your browser (usually port 8000 or similar).

3) Ingest Your Docs with Regolo Embeddings

We need to turn your PDFs/MDs into vectors. Use the specialized Regolo embedder.

git clone https://github.com/regolo-ai/regolo-docs-embedder
cd regolo-docs-embedder
# Follow README to point to your local docs folder
python embed.py --api_key $REGOLO_API_KEYCode language: Bash (bash)

Expected output: Your local Weaviate is now populated with vectors generated by Qwen3-Embedding-8B (via Regolo).

4) Connect Elysia to Regolo (The Brain)

In the Elysia UI Settings, configure the “Brain” to use Regolo’s inference.

  • Provider: openai (Regolo is compatible)
  • Base URL: https://api.regolo.ai/v1
  • API Key: YOUR_REGOLO_KEY
  • Base Model: Qwen2.5-14B-Instruct (Fast, for simple routing)
  • Complex Model: Llama-3.3-70B-Instruct or gpt-oss-120b (Smart, for reasoning)

5) Chat & Reason

Ask a complex question: “Analyze the Q3 financial reports and summarize the risk factors in a table.”
Expected output: Elysia uses the “Complex Model” to plan the task, queries Weaviate for Q3 docs, and renders a structured table in the UI—not just a markdown block.

Production-Ready: Agent Configuration

To make the agent consistent, tune the system prompt in Elysia’s config file (elysia_config.yaml or UI).

agent:
  system_prompt: |
    You are a Senior Analyst. 
    ALWAYS cite your sources using [Doc ID].
    If the user asks for data comparison, output a JSON that the UI renders as a table.
  tools:
    - name: "weaviate_search"
      description: "Search the vector DB"
    - name: "calculator"
      description: "Perform math on retrieved numbers"Code language: Bash (bash)

Benchmarks & Costs

FeatureRegolo + Elysia (Self-Hosted)SaaS RAG (ChatPDF, etc.)
PrivacyHigh. Docs local, inference stateless.Low. Docs uploaded to their cloud.
IntelligenceAgentic. Plans multi-step answers.Single-step retrieval only.
CostInference only (~€5/mo for heavy use).Subscription fees (€20+/user).
UIDynamic. Tables, charts, cards.Text only.

👉Deploy your Second Brain

Resources & Community

Official Documentation:

  • Regolo Platform – European LLM provider, Zero Data-Retention and 100% Green

Related Guides:

Join the Community:

🚀 Ready to Deploy?

Get Free Regolo Credits →

Built with ❤️ by the Regolo team. Questions? support@regolo.ai