Research RAG

Your papers. Your answers. Every claim cited.

Research RAG indexes your PDFs, notes and GitHub repos and answers questions across all of them - with every reply traced back to the exact page or line it came from. Private by default, never used for training.

Get Research RAG Use it from the terminal

How Research RAG works

Five steps from your file to a cited answer. Nothing leaves your workspace.

01ConnectAdd PDFs, arXiv links, plain-text notes or GitHub repo URLs. Any format, any source.

02ParseEquations, tables, figures, code blocks and section structure are all preserved, not flattened.

03IndexDual index: dense vector search for semantic similarity, BM25 keyword search for exact terms. Both run on every query.

04RetrievePassages ranked by relevance, recency and citation weight. Filtered for prompt injection before they reach the model.

05AnswerEvery claim in the answer carries a citation - paper title, section, page or line number. No confident guessing.

What you can add

Research RAG accepts anything a researcher might work from.

PDFs

Papers, reports, theses. Equations and tables parsed structurally, not as raw text.

kx: kx rag add paper.pdf or kx rag add ./papers/ --recursive

arXiv links

Paste an arXiv URL or ID. Research RAG fetches the PDF and indexes it automatically.

kx: kx rag add arxiv:2205.14135

Notes and text files

Markdown notes, reading summaries, annotation exports. Ask across your own annotations alongside the papers they annotate.

kx: kx rag add notes/flash-attn.md

GitHub repos

Add a repo URL and Research RAG indexes the code, README and any markdown docs. Ask about the implementation alongside the paper it implements.

kx: kx rag add https://github.com/user/repo

Paper Lab papers

Any paper in your Paper Lab bucket is already in your Research RAG. No upload needed - it stays in sync.

kx: kx rag ask "What architecture does FlashAttention use?"

Researcher catalogs

Search a researcher by ORCID or Google Scholar and add their full publication list to your RAG in one click from the Researcher search panel.

kx: kx scholar "Tri Dao" --add-rag

Live catalog

4 papers in the Research RAG catalog

These papers are pre-indexed and available to all Research RAG users. Your private files sit alongside them, visible only to you.

Transformer Free demo NeurIPS 2017

Open in Paper Lab

Attention Is All You Need

Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, Illia Polosukhin · arXiv:1706.03762

attentiontransformersequence-modellingnlparchitecture

FlashAttention NeurIPS 2022

Open in Paper Lab

FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness

Tri Dao, Daniel Y. Fu, Stefano Ermon, Atri Rudra, Christopher Re · arXiv:2205.14135

attentiongpumemory-efficiencyio-awarenesscudasystems

TriCalRAG Preprint 2026

Open in Paper Lab

TriCalRAG: A Three-Strategy, Retrieval-Augmented Bench for On-Prem LLM-Based Root Cause Analysis in AIOps

Rohit S Patel, Subhrakanta Mohanty, Jeenal Chaudhary · arXiv:2609.14762

ragaiopsroot-cause-analysison-premisellmretrieval

TriShieldRAG Preprint 2026

Open in Paper Lab

TriShieldRAG: 3 Rings, One Blind Spot in Layered Defenses for RAG

Subhrakanta Mohanty, Rohit S Patel, Yuvaraj, Jeenal Chaudhary, Singhania · arXiv:2607.23838

ragsecurityadversarialprompt-injectiondefenseretrieval

Topic clusters

Papers grouped by research theme. Search within a cluster for focused retrieval.

Attention and Transformers

Self-attention, multi-head attention, efficient attention variants and the Transformer family.

2 papers: Transformer FlashAttention

Retrieval-Augmented Generation

RAG architectures, benchmarks, defense against poisoning and root cause analysis applications.

2 papers: TriCalRAG TriShieldRAG

GPU and Systems Efficiency

IO-aware kernels, memory-efficient inference, CUDA optimization and hardware-aware algorithm design.

1 paper: FlashAttention

AIOps and Observability

LLM-assisted operations, root cause analysis, on-premises deployment and reliability engineering.

1 paper: TriCalRAG

KXAI + Research RAG

Ask KXAI across your whole library, not just the demo.

KXAI in demo mode is scoped to papers in the shared bucket. Upgrade to Scholar and KXAI answers from everything you've added - your uploads, your notes, your repos - with every answer still cited back to the source.

Demo mode (free)

Grounded Q&A on 4 shared demo papers
Answers cited to Architecture, Cheatsheet, Results tabs
Cross-paper search across the shared bucket
No account needed

Scholar and Professional (your RAG)

Everything above, plus your own uploaded papers
Ask across PDFs, notes, GitHub repos and arXiv links you added
Add a researcher's full paper list from Researcher search in one click
Cross-paper reasoning - answers synthesised across two or more papers
Private: your files are never used for training

Terminal access

Research RAG from the kx CLI.

Every Research RAG operation is available from the terminal. Add files, ask questions, list your library - all pipeable to JSON for scripts and CI.

kx rag add ./papers/ --recursive Add every PDF in a folder to your Research RAG in one command.

Also accepts: single PDFs, arXiv IDs, GitHub URLs, plain-text notes.

kx rag ask "How did we handle fault injection?" Ask across your private library. Every answer cites the exact page or line.

Pipe: kx rag ask "..." --json | jq '.answer'

kx scholar "Tri Dao" --add-rag Look up a researcher by name or ORCID and add all their public papers to your RAG at once.

Also: kx rag list --json | jq '.[].name' to see what's indexed.

Full CLI reference

Safe retrieval by default

Research RAG is designed for private research. Three properties hold on every query.

Private by default

Your uploaded files stay in your workspace. They are never shared with other users and never used to train our models.

Injection-filtered

Retrieved passages are checked for prompt injection before they reach the model. Based on the approach in our TriShieldRAG research.

Always cited

Every claim in every answer links back to a source. If the answer isn't in your indexed files, Research RAG says so rather than guessing.

Common questions

Is my Research RAG shared with anyone? No. Your uploaded files and the index built from them are private to your account. Team workspaces on Professional plans share within the team only, with role-based access control.

Can I add papers from a specific researcher? Yes. Use Researcher search on the /ai/ page to find a researcher by ORCID, Google Scholar link or name, then click "Add to RAG" on any of their papers. From the terminal: kx scholar "name" --add-rag.

What happens to equations and code? They are parsed structurally and kept intact. LaTeX equations index as equations, not as flattened text. Code blocks index as code. You can ask about an equation by name or ask for a function by description.

How is this different from just using a general LLM? A general LLM answers from training data and cannot tell you which paper a claim came from. Research RAG answers only from the files you indexed, with every claim cited to the exact source page or line. No hallucination from prior training.

What plan do I need? Research RAG is a Scholar and Professional feature. Scholar includes about 2 GB of private RAG storage. Professional includes about 20 GB plus team workspace support and CLI access.