DEV Community

Cover image for Building Igris: An Autonomous AI Agent with "No-Vector-DB" RAG
Abhishek Dvs
Abhishek Dvs

Posted on

Building Igris: An Autonomous AI Agent with "No-Vector-DB" RAG

When I started building Igris, I had one goal: move beyond the "chat window." I didn't want a wrapper; I wanted an Autonomous Execution Agent. Igris is designed to be my digital shadow—capable of answering for me, acting as me, and executing tasks end-to-end.

In this post, I’ll break down the full-stack architecture, the "free-tier" DevOps hacks, and how I implemented a personality-driven knowledge base without the overhead of a vector database.

🛠 The Tech Stack
I chose a stack that emphasizes speed, reliability, and end-to-end control:

Backend: Node.js + Express.

AI Orchestration: LangChain with a custom fallback architecture

Frontend: A dual-approach:

React + Vite for the standalone, high-performance chat interface.

Astro for seamless integration into my personal portfolio.

Hosting: Render (API) and Cloudflare Pages (Frontend).

🏗 Key Features & Architecture

  1. The "Shadow" Personality Instead of generic system prompts, Igris uses specific context files:

One contains everything about my professional background, projects, and tech stack. This allows Igris to act as my "Shadow" and answer inquiries as if I were there.

Another - defines the agent's internal persona, tone, and decision-making logic.

  1. Lightweight RAG (Knowledge Base)
    Most tutorials say you need a vector database for RAG. I took a different route for the initial build. By leveraging a structured Knowledge Base and injecting relevant context directly from our file system into the prompt window, I achieved high accuracy and low latency without the complexity of an external DB.

  2. Reliability & Guardrails
    I implemented a robust fallback logic in my agents.js. If the primary LLM provider fails, Igris automatically switches to a alternative provider. This is coupled with strict guardrails to ensure the agent stays within its execution scope.

🚀 DevOps on a Budget: The Render Hack
Hosting on Render's free tier comes with the "spin-down" problem. To keep Igris responsive 24/7 without a credit card on file, I implemented Cron Jobs. This ensures the Node.js instance stays warm and ready for execution.

🔗 Check it out
Live Demo: Agent Igris Standalone

Portfolio: abhistack.pages.dev

Building this taught me that the "Agentic" part of AI isn't about the model—it's about the orchestration around it. Questions? Drop a comment!

Top comments (0)