SapienAI

What SapienAI gets right is context. Every other research stack forces you to re-explain your project to each app — the writing tool doesn't know your sources, the reference manager doesn't know your draft, the notebook doesn't know either. SapienAI's Research Spaces hold project files, collections, graph views, agent sessions, collaborators and layouts in one place, and universal search reaches across files, collections, spaces, your Zotero library and prior agent sessions. That means a literature review or a paper moves from first source to final draft without leaving the project. The capability list is unusually long for a research tool, and it is the specific mechanics that matter. An agent pointed at your library reports its own progress — "Searched 142 sources, read and ranked 24 passages, synthesising findings" — and stops for approval. Real-time collaborative editing runs across LaTeX, Typst and Markdown in the same surface, with live cursors. Python, R and Jupyter execute in the browser with no install, and figures, tables and outputs save back to the project and stay citable. The Knowledge Explorer shows entities, claims, topics, files and source segments in graph and card views. The evidence-native wiki turns a collection into linked, versioned pages that preserve citations, history and disagreement. The AI dial is the detail most competitors miss. SapienAI is a complete research workspace before a single token is generated — real-time collaborative editing, Python/R/Jupyter computation, knowledge graph and semantic search all run with GenAI off. You choose Off, Assist or Full agents. On the self-hosted version you also choose which model powers a chat — the docs name GPT-4, o3, Claude 4, Google Gemini 2.5 Pro and Ollama, with support for changing models mid-chat, routing through Azure, AWS or Google Vertex, and realtime audio chat. Model hosting flexibility of that kind is rare in academic tooling. Where it falls short is readiness, not capability. The pricing page says outright that pricing is being finalised and directs you to the waitlist for details on individual, institutional and self-hosted options. The vendor's own beta notice says SapienAI is under active development, testing has been limited and only on a small range of devices, and final features, availability and pricing are still under review, with some capabilities described as planned. Access for the hosted route is a waitlist. The self-hosted route asks you to operate and maintain your own deployment and configure it around your technical environment. So the honest read: SapienAI is a strong fit for a lab or department with someone who can run a self-hosted deployment and wants AI grounded in real papers rather than a generic chatbot. It is a poor fit if you need a managed service with a confirmed price this quarter, or if nobody on your team wants to own infrastructure. Treat the current build as a serious, documented beta — the docs are public, the GitHub repo