Aktilot
Build AI Agents that understands your documents.
8 followers
Build AI Agents that understands your documents.
8 followers
Aktilot is a self-hosted, open-source RAG platform for chatting with your documents. Hybrid BM25 + vector retrieval, durable Temporal workflows, full observability, and cited answers — all running on your own infrastructure. No data leaves your servers.
👋 Hi everyone!
I'm Vikas, the creator of Aktilot. (https://github.com/vikas0686/Aktilot)
I started building Aktilot because I realized that building a production AI application involves much more than just connecting an LLM to a vector database. You need document ingestion, hybrid search, reranking, workflow orchestration, benchmarking, observability, and evaluation—all working together.
Aktilot is an open-source AI platform that helps developers build, evaluate, and operate production-ready AI applications.
What you can do with Aktilot
📄 Chat with your documents using RAG
🔍 Hybrid search (vector + keyword)
🔄 Workflow orchestration for AI pipelines
📈 Built-in observability and execution insights
🤖 Support for both cloud and local LLMs (including Ollama)
My goal is simple: make it easier for developers to move from AI prototypes to production systems.
Aktilot is completely open source, and I'd love your feedback.
A few questions for the community:
- What's been the biggest challenge in taking an AI application to production?
- Which feature would make Aktilot most valuable for your team?
Thanks for checking it out! I'll be here throughout the day to answer questions and discuss ideas. 🚀
The choice to lean on Temporal for durability instead of stitching together cron jobs or a fragile queue shows real thought about what breaks in production. Self-hosted RAG that doesn't cut corners on observability is rare.
The hybrid retrieval setup sounds solid, especially for teams that need answers grounded in their own docs. One thing I'd love to see is a built-in evaluation harness so you can quickly benchmark retrieval quality against your own golden Q&A sets after tweaking chunk sizes or embedding models.