Kotaemon: 24K Stars Can't Fix These RAG Problems
Kotaemon promises document chat without ML expertise—11K stars in 9 months prove developers want this. But 106 open issues reveal the gap between prototype and production. Installation breaks, processing crawls, and API limits hit fast. Here's what actually works, what breaks, and what it teaches about making RAG tools production-ready.
A six-page resume takes 30 minutes to process. OpenAI's API starts refusing requests because the app spawns too many threads. Docker builds break on fresh installs. KeyErrors crash the GraphRAG integration mid-query.
This is Kotaemon in production—or rather, trying to reach production. The open-source RAG framework promised document Q&A without hiring ML engineers, and 24,000 GitHub stars suggest developers want that. But 106 open issues tell a different story about the gap between spinning up a prototype and shipping something customers can rely on.
The Promise: Document Chat Without the ML Degree
Kotaemon delivers technical sophistication out of the box. Hybrid RAG combines full-text search, vector retrieval, and re-ranking in a single pipeline. Multi-modal support extracts answers from figures and tables, not just text. ReAct and ReWOO agents handle complex multi-step queries. The Gradio-based UI shows PDF citations directly in-browser, letting users verify answers without juggling windows.
For internal tools teams, this matters. You can clone the repo, point it at a folder of PDFs, and have a working document chat interface running locally—no ML degree required. The 24,000+ stars reflect real demand: backend developers who need RAG capabilities but can't justify enterprise contracts.
The Reality Check: When 6 Pages Take 30 Minutes
GitHub issue trackers don't lie. In August 2024, Kotaemon had 20 open issues—mostly file handling bugs and installation friction for non-technical users. By late 2025, that number hit 106. The pattern reveals where "accessible AI" breaks down.
Processing speeds collapse under modest loads. One user reported 30 minutes to index a six-page resume. Thread management spirals out of control, triggering OpenAI API rate limits as the app spawns excessive concurrent requests. The NanoGraphRag integration throws KeyErrors during queries, breaking the knowledge graph features that differentiate Kotaemon from simpler vector-only tools.
These aren't coding mistakes—they're the engineering realities of making complex systems simple. Managing thread pools, handling API backpressure, and stabilizing graph database integrations are hard problems that don't disappear because you wrapped them in a friendly UI.
Installation Roulette: Dependency Hell Never Sleeps
Packaging AI tools means wrangling dependencies across Python, Node, and Docker. What works on the maintainer's M1 Mac breaks on Linux servers. Conda environments conflict with system Python. Docker builds fail on certain kernel versions.
The installation issues flooding GitHub aren't documentation failures—they're the reality of shipping software that depends on transformers, vector databases, OCR libraries, and graph engines simultaneously. Each dependency brings its own compatibility matrix. Multiply that across operating systems and Python versions, and "just works" becomes a statistical improbability.
What Kotaemon Gets Right
Kotaemon's hybrid retrieval shows measurable advantages for document-heavy use cases. Full-text search catches exact matches that embeddings miss. The in-browser citation preview beats separate windows for UX flow.
The Gradio foundation makes customization straightforward for Python developers. You can modify the UI, swap retrieval strategies, or add custom preprocessing without forking the entire codebase. For internal tools where you control both the data and the users, this flexibility matters more than polish.
The Verdict: Who Should Use Kotaemon Today
Kotaemon works for internal prototypes with under 50 documents and technical users who can debug Python stack traces. It's the wrong choice for customer-facing features or anything non-technical end users need to install themselves.
The 24,000 stars demonstrate unmet demand for accessible RAG tooling. The 106 open issues—and the trajectory from 20 to 106 in months—show we're still early in solving the production-readiness problem. Kotaemon proves developers can build sophisticated document Q&A without ML teams. It also proves that "build it yourself" still means "debug it yourself" when processing slows, APIs fail, or dependencies conflict.
Making AI accessible means more than lowering the barrier to Hello World. It means handling the thousand small failures that happen between prototype and production.
Cinnamon/kotaemon
An open-source RAG-based tool for chatting with your documents.