You’ve probably felt the frustration. You feed an AI your notes, PDFs, project docs, or Obsidian vault — ask it something important the next day — and it replies like it’s never seen your stuff before. Or you paste the same context over and over and watch the token count explode.
That problem is now solved.
KEY RELEASE: Khoj (khoj-ai/khoj) hit #1 trending on GitHub in Python Agents with 32.8K stars. It’s a fully open-source, self-hosted AI second brain that indexes your personal files, provides lightning-fast semantic search, builds custom agents, schedules automations in plain English, and runs deep multi-step research — all while keeping every byte of your data on your machine.
What It Actually Does
Khoj is your private AI that lives on your Mac, Linux box, or server. It indexes and understands all your personal files — PDFs, Markdown, Notion exports, Word docs, Org-mode, Obsidian vaults — with lightning-fast semantic search. You chat naturally about anything, and it pulls exactly the right context from your docs or the web without you copying and pasting a thing.
But it goes well beyond search. You can build custom agents with their own personality, tools, and knowledge bases — a \"Research Agent,\" a \"Meeting Summarizer,\" a \"Code Reviewer.\" You can schedule automations in plain English: \"every Monday morning generate a weekly newsletter from my notes and email it to me.\" Start any message with /research for multi-step deep thinking with tool use. And it works with any LLM you already love — local (Llama 3, Qwen, Mistral, Gemma) or cloud (GPT, Claude, Gemini, DeepSeek).
Why This Is a Big Deal
Most personal AI tools are either cloud-only (your data leaves your computer) or forgetful one-shot chatbots that don’t remember week-to-week. Khoj is different. It’s fully open-source (AGPLv3), runs 100% locally if you want, and feels like the persistent, lifelong agent we’ve been waiting for. It’s perfect for heavy local users on Mac (pairs beautifully with Ollama or MLX) and instantly makes your existing agents 10× more useful because they finally have real memory and your personal context.
At TrainingRun.AI we’re already spinning up Khoj instances for our own workflows and will be adding it to TRAgents scoring next week — especially in the Task Completion and Multi-Model Efficiency pillars. Persistent personal agents with real memory and custom tools are exactly what moves the needle on real-world daily use.
Your data never has to leave your private network. You can even use Khoj without an internet connection.
Quick Start
Two minutes on Mac or Linux:
git clone https://github.com/khoj-ai/khoj.git
cd khoj
docker-compose up -dThen open http://localhost:42110, point it at your Obsidian/Notion/PDF folder, pick your LLM, and start chatting. Full setup docs at docs.khoj.dev.
This is the kind of practical, privacy-first release that makes 2026 feel like the year personal AI finally grows up.
GitHub (32.8K stars and climbing): github.com/khoj-ai/khoj · Live cloud demo: app.khoj.dev · Docs: docs.khoj.dev
Check out the project: Khoj (khoj-ai/khoj)
View on GitHub →What do you think? Drop a reply on X. We read every one.
— The TrainingRun.AI Team
