Academic Research & Scholarly Intelligence, 100% On-Device.
Read complex scientific manuscripts with instant AI explanations, run dense semantic vector retrieval via in-browser PGlite, and draft literature matrices completely private on your device with Google LiteRT.
Zero cloud API calls, running Gemma 4 E2B locally on edge hardware.
Full PostgreSQL + pgvector compiled to WASM for instant dense retrieval.
Unpublished manuscripts never leave your device. Complete offline security.
Private by Architecture. Powerful by Design.
Buddhi AI shifts the computational burden entirely from remote cloud clusters to client-side hardware. Zero server computation costs and zero external data transmission.
Google LiteRT Runtime
Native WebGPU hardware acceleration using @litert-lm/core and @litertjs/core. Runs Gemma 4 E2B and EmbeddingGemma 300M directly in edge browser memory without remote inference calls.
PGlite Vector Store (WASM)
Full-featured PostgreSQL compiled to WebAssembly via @electric-sql/pglite with pgvector support. Executes dense vector similarity queries, document chunk caching, and relational filters right in browser memory.
RLM & In-Browser Sandbox
Recursive Language Models (RLM) deconstruct multi-page academic papers into structured hierarchies. Integrated with @buddhilive/sandbox to compile and live-preview Next.js applications client-side via Service Worker routing.
From Raw Manuscript to Synthesized Literature Review
Every stage in the research lifecycle executes locally on your device with complete cryptographic isolation.
Document Ingestion
Upload PDF or LaTeX papers. Extracts text, tables, and frontmatter metadata locally via pdfjs-dist.
Vector Embeddings
EmbeddingGemma 300M generates dense 2048-dim vectors into WASM PGlite with zero cloud latency.
RLM Decomposition
Recursive language models summarize methodology sections, isolate experimental limits, and map cross-paper concepts.
Citation & Drafting
Draft conference abstracts, generate verified APA/IEEE citations, and export literature review matrices.