Mugib
AI Agent Platform for customer support,
4 followers
AI Agent Platform for customer support,
4 followers
Your business has knowledge. Mugib turns it into an AI agent that works 24/7. Upload your docs, connect your live data sources, and deploy instantly across WhatsApp, Telegram, Instagram, Email & more. Real answers from your real data — not generic replies. No coding. No per-conversation pricing. Your data stays yours. 🇮🇶 Built in Iraq, for the world.
Free Options
Launch Team
Framer AI AgentsDesign and publish professional sites with AI
Promoted
Great to see another knowledge-base AI agent platform. About your embedding approach, are you doing chunk-level retrieval or full-doc context? I'm building something similar focused on web agencies and the RAG pipeline choices made a huge difference for answer quality.
@cuygun
Hey Cesurhan! Appreciate that 🙌
RAG is just one piece of what we built honestly. The part that made the biggest difference for us was the routing layer — the agent decides which source to hit before even touching the knowledge base. Sometimes the answer isn't in a document, it's in a live database or an external API, so we built MCP integration to pull real-time data directly into the response.
On top of that we have context memory across sessions, multi-LLM failover so the agent never goes down, and safety guardrails that keep responses scoped strictly to your data — no hallucinations from general training.
The RAG itself we tuned heavily too, but the combination of all these layers is what made answer quality jump for us.
What's your pipeline looking like so far?
@anas_abdullkarim Nice, the routing layer before RAG makes a lot of sense. We went a different direction, chunk-level retrieval with cosine similarity on pgvector, then a reranking step that weighs recency and source type. Added a curated Q&A layer on top so clients can pin exact answers for high-stakes questions (pricing, policies) without relying on retrieval at all.
Multi-LLM failover is smart though, we've been single-provider so far but that's on the list. How are you handling the context window when you're pulling from both docs and live APIs in the same turn?