SLM turns any corpus — a book, a policy library, an operations manual — into a specialized model that answers only from your material, with every claim verified and cited.
$ make train PACK=operations ✓ 503 sections indexed ✓ hybrid index built (bm25 + dense + FAISS) ✓ adapter lora_3 ready You: What does our policy require for access? SLM: Per Section 4.2 (p.41): manager approval within 72 hours. [cite → 4.2 · page 41]
The gap
Ask a general-purpose LLM about your internal material and it improvises — blending memory, hearsay, and confidence into prose you cannot verify.
General models assert facts they never saw in your documents, with no way to tell which claim is real.
No citations, no page numbers, no path back to the source. You cannot audit what it told you.
You cannot scope what it knows, restrict access, or guarantee it answers from the right edition.
What changes for you
PDFs, Word, Excel, PPT, Markdown, plain text, URLs. SLM ingests, structures, and indexes the material.
A single "Train" button builds the pack — chunking, retrieval index, and adapter — with zero ML knowledge required.
Every answer is retrieved from your material, verified claim-by-claim, and returned with a citation you can open.
One engine, three ways to use it
Ask anything about a book, spec, or manual. Answers grounded in the text and cited to the exact section.
Which passage supports this? →Train on your policy library. Ask what the rules say — and get the exact clause, section, and page.
What does our policy require here? →Feed it runbooks and procedures. Get step-by-step answers grounded in your actual operating material.
Walk me through the runbook →How it works
Extract text with structure intact — chapters, sections, footnotes — and clean headers, footers, and OCR noise.
A hybrid index (lexical + semantic + FAISS) makes every passage retrievable; claim verification checks each sentence against the source.
Questions are answered from retrieved passages, every claim verified, unsupported statements dropped — never invented.
You: Who is Anjaneya? SLM: Son of the God of Wind, by name Hanumán. [cite → Canto LI · Hanumán's Reply] You: What is the capital of France per this text? SLM: That is not stated in this material. (verified abstention — no source, no guess)
Deployment
Workstation, single VM, or air-gapped — one engine, every posture. The engine is corpus-agnostic: add a pack, swap the expertise, zero code changes.
Live now
The public console is grounded in three real corpora. Ask about the epic, the novella, or the synthetic Atterby corpus — every answer is cited.
Get started
Download SLM, point it at a real corpus, and see a grounded, cited expert in under thirty minutes.