Tools for districts, open datasets, trained models, papers, SDKs and the Learnlog draft standard. Each item shows its license. Free tools that aren't open source say so.
Ready to use with a team this week. No sign-in.
A 40-minute session for teacher teams grading on proficiency scales. It ends in two dated agreement percentages.
CC BY 4.0One page for deciding how student support counts toward a proficiency decision.
CC BY 4.0Exact and adjacent agreement from a norming grid. It makes no network calls and works offline, so your data stays on your computer.
Free to useFive questions every district should ask every education-AI vendor, with what a good answer looks like.
CC BY 4.0A three-page brief on why open education AI needs a shared, checkable way to measure what works.
Free to readA four-page guide to assignments where the process is the evidence, so using AI to skip the thinking doesn't work.
Free to readFree to use. These are QLM products, not open source.
Interactive science and math worlds, by grade band and subject. Students predict and experiment, and teachers see misconception evidence.
Free to usePractice teaching with virtual students and get coaching on each session. Five free sessions a day.
Free to useAI patients with gated histories: you draw out the facts by asking. Optometry and nursing programs.
Free to useCredentials that record what a student demonstrated, not just what they completed. Programs in clinical optometry, clinical nursing and general medicine.
Free to useMeasurement infrastructure for researchers and districts: run studies, share results others can check, and calibrate content.
Free to useFifteen real questions for the SAT, ACT or AP. No sign-up. You get an estimated score, the gaps, and a study plan.
Free to useDownload them, cite them, build on them.
423 research-cited mathematics misconceptions, K-16, built for the LabPath worlds.
CC BY 4.0The learning graph, NGSS standards alignment, and a crosswalk that maps QLM misconceptions to other datasets.
CC BY 4.0Browse the ontology by Common Core math standard.
Free to useTwo calibrations on public ASSISTments data, each with its report and parameter file, plus synthetic recovery checks, labeled as synthetic.
Free to readOn Hugging Face. Each model card leads with its limitations; read them before you use a model.
Labels a K-12 math student response with the misconception it shows, from 39 grouped categories. It catches most misconceptions, and about half of what it flags is wrong.
MITLabels an utterance from K-12 math classroom dialogue with the teaching talk move it performs.
CC BY-NC-SA 4.0 · non-commercialAn 8B math tutoring model built on Llama 3.1 8B Instruct. Access is gated: request it on Hugging Face.
Llama 3.1 license · gatedOpen source. Install it, read it, change it.
Evidence events, record verification and dataset tools for education AI, in JavaScript and Python.
Apache-2.0Turns the logs you already have into Learnlog records. Adapters for TutorMoments, OATutor, Sandpiper (the National Tutoring Observatory's annotation app) and an explicit help-event format, plus an Apache Airflow pipeline.
Apache-2.0Ask your AI assistant what learning records actually show about help. Eight read-only tools and four Claude skills.
Apache-2.0Drop in a measurement record and check it. The page can't make network requests, so nothing leaves your device.
Apache-2.0A kernel for studies with an LLM in the loop that others can rerun and check.
Apache-2.0Learnlog is a proposed open format for recording AI help in learning. It keeps each stage of help apart, and it never turns "not recorded" into "none."
An individual Internet-Draft, revision 03. It is a proposal, not an approved IETF standard.
IETF TrustPreprints, a standards paper, and preregistrations, newest first.