Private by construction
With a local model, your book, your answers, and your voice never leave the machine. Nothing to leak, nothing to retain.
Local-first AI applications
Capable AI that stays on your machine.
JMNI Labs builds AI applications that run on hardware you own — no data sent to someone else's cloud. Our first is Recitation, a private AI professor that turns any textbook into a complete course.
Runs with
OllamaLM StudiovLLMllama.cppAny OpenAI-compatible API
Recitation
Not a chatbot. A course.
Most AI tutors answer questions. Recitation teaches: it reads your book, designs the syllabus, lectures section by section, assigns problems, grades them — and remembers where you left off.
Upload your book
Any textbook — PDF, text, or Markdown. Chapters, tables, and figures are indexed on your machine.
Get a curriculum
Units, lessons, and objectives built from the book's own structure, paced to your level and prior courses.
Learn interactively
Streamed lectures that stop for questions, check comprehension, and cite the pages they draw from.
Prove it
Graded problem sets with hints and explanations. Mastered concepts are tracked across every session.
Why local-first
Your material is yours. It should stay that way.
Textbooks, lecture notes, internal manuals, student records. The things people most need to learn from are often the things they can't — or shouldn't — upload to a third party.
Grounded, not guessed
Answers come from your book first, with page citations. The professor is built never to invent a source.
Any model you choose
A laptop running Ollama, a server down the hall, or a cloud API you trust. Weak local models are supported gracefully.
Yours to keep
No per-seat meter for individuals, no deprecations, no terms-of-service surprises. It keeps working offline.
Who it's for
Built for people who study seriously.
Independent learners
Work through the textbook you always meant to finish — with a professor who never runs out of office hours.
PERSONAL · FREE
Certification & exam prep
Turn a canonical review text into a structured course with problem sets that find your weak spots.
PERSONAL · FREE
Schools & universities
AI tutoring on hardware your institution controls, so student data and course materials stay on campus.
INSTITUTIONS · DESIGN PARTNERS
Teams & organizations
Train staff on internal manuals, procedures, and policies — documents that should never touch a public AI service.
TEAMS · EARLY ACCESS
JMNI Research
We measure what local AI can really do.
Behind the products is our research lab: six NVIDIA GB10 systems where we run frontier open-weight models across switchless multi-node rings — and record every result.
It keeps us honest. Recitation is designed around what local models actually deliver, not what a cloud API promises.
1Mtokenscontext retrieval benchmarked on a four-node ring
8 × 258Kconcurrent long-context sessions verified on two nodes
~2,400tok/sprefill on a single GB10 system
0network switches in the model path
Measured on the JMNI Research fleet. Methodology and limits →
Field Notes
From the lab.
Bring a textbook. Leave with a course.
Recitation is in private preview. We're onboarding individual learners, and a small number of schools and teams as design partners.