Abstimmbar is an open-source audience response system (ARS): lecturers run live quizzes and polls in class, students answer anonymously on their own devices via QR code or short URL — no login, no app. It integrates into learning management systems via LTI 1.3 / LTI Advantage and supports staff login via OpenID Connect (e.g. Keycloak).
It is the successor to the Stud.IP plugin Cliqr, feature-wise oriented at ARSnova/Particify, and is developed at virtUOS, Universität Osnabrück as a sibling project to Ausleihbar, sharing its design language and stack. Released under the Apache License 2.0.
Status: usable. The full MVP loop (M0–M3), LTI 1.3 integration (M4) and a broad set of v2 features are implemented and load-tested at 1000 concurrent participants. Pending: manual acceptance against a production LMS. Reviewable background documents:
docs/concept.md— Funktionsumfang & fachliches Konzeptdocs/roadmap.md— Meilensteine (MVP → v2 → Ausblick)docs/decisions/— Architekturentscheidungen (ADRs)docs/anleitung-lehrende.md— Anleitung für Lehrendedocs/deployment.md·docs/monitoring.md— Betrieb und Monitoring (Prometheus/Grafana)
Presenter view — a live question with real-time results on the beamer.
- 🎓 Teacher-paced live quizzes — start/stop each question from a distraction-free presenter view (beamer-friendly, keyboard shortcuts), live vote counter, results as bar charts.
- 🗂️ Three set types, each with its own purpose — chosen at creation and
fixed afterwards, the type decides how a set is run and which question
formats it allows:
- Live poll — presenter-driven on the beamer, you start and stop each question; all question types.
- Self-paced quiz — participants work through the set at their own pace in class; you start it and watch the results come in. All question types.
- Self-check — learners practise on their own via a standing link with immediate feedback (auto-checkable formats: single/multiple choice, ordering, open text).
- 📱 Anonymous participation — join via QR code, short URL or room code; an ultra-lightweight, framework-free participant page that loads instantly on phones in a packed lecture hall. No account, no IP logging on votes.
- 🧩 Question formats — single & multiple choice (with images as options), word clouds, Likert scales, open text, priorities and ordering/ranking. Optional per-question countdown timer; optional section slides to structure a set.
- 🔗 LMS integration via LTI 1.3 — resource-link launch and deep linking (Stud.IP, Moodle, ILIAS, …), embeddable in an LMS iframe; no LTI 1.1 legacy. Grade passback (AGS/NRPS) planned for assessed use later.
- 🔐 University SSO — OIDC login (Keycloak) for lecturers and admins, with back-channel logout. Any authenticated person can create rooms.
- 🌐 Bilingual DE/EN — both the interface and authored content (per-language question and option text), with an optional machine- translation pre-fill in the editor (see below).
- 📊 Results & reuse — results view with per-run deletion and CSV export, sanitized JSON export/import, question-set duplication across rooms, full-text search, sharing & co-ownership of rooms and sets.
- 🖼️ Robust image handling — drag-and-drop images in questions are normalized on upload (downscaled, re-encoded to WebP) so they stay sharp on beamer and phone without bloating storage.
- 📈 Usage statistics & monitoring — a staff-only statistics page in the
admin area shows totals (rooms, users, sets & questions created and
conducted, participants, guided-tour usage), breakdowns by type as donut
charts and time series over a selectable date range. An optional,
token-guarded Prometheus
/metricsendpoint feeds the same figures into Grafana (setup:docs/monitoring.md). Only aggregated, anonymous data — no per-user tracking; session keys are hashed and pruned; example rooms are excluded from usage figures; tour usage is stored only as daily counters.
Joining is instant and anonymous — students scan a QR code or type a short code; the participant page is a tiny, framework-free bundle that loads fast on phones in a packed lecture hall.
 |
 |
Author once, present anywhere — a lean WYSIWYG editor with seven question types, sections, drag-and-drop images and per-language DE/EN content.
 |
 |
Results during and after the session — live bar charts and word clouds in the presenter view; a stored results page afterwards with CSV export, plus a self-paced quiz mode students can work through on their own.
Live word cloud — terms sized by frequency (case and spelling variants merged).
 |
 |
The whole interface is theme-aware (light / dark) and bilingual DE/EN.
Abstimmbar ships a set of opt-in AI helpers that assist authoring and analysis without ever taking a human out of the loop. They are disabled out of the box and only activate when a deployment points them at an LLM endpoint — which can be a self-hosted, OpenAI-compatible model (e.g. via a LiteLLM proxy), so student data never has to leave your institution. Every AI output is a draft a human reviews before anything is saved, and vote tallies are always recomputed from the real votes — the model can group and label, never invent numbers.
Generate draft questions from a slide deck — you choose which to keep.
- 🧠 Generate questions from your slides — drag-and-drop (or pick) a PDF, PPTX or ODP, or paste text, and get draft questions across all supported types. Coverage spans the whole document; you set the density (questions per page) and cognitive level (Bloom-inspired), and can prioritise one question type. Large documents are processed as a resumable background job with a live progress indicator, then you review the drafts and pick which to keep — nothing is imported automatically.
- ✏️ Authoring assists — one click to suggest plausible distractors for a choice question, or to rephrase a question more clearly. Suggestions are proposed, never auto-applied.
- ☁️ Smarter word clouds — during a run, the AI view merges spelling variants, typos and synonyms into groups and sorts them into themes (automatic, or by a criterion you set, e.g. "by music genre"). Computed live and kept warm so the view is instant; counts come from the raw votes.
- 📝 Open-text evaluation — classify free-text answers into categories (default correct / unclear / wrong, or your own 2–5 labels) against an optional reference answer, live as votes arrive or on demand for a whole run.
- 📄 Run summaries — generate a short, display-only Markdown summary of a run's aggregated, anonymous results.
- 🌍 Machine translation — optional pre-fill of DE/EN content translations in the editor via a self-hosted LibreTranslate; it drafts into empty fields and never overwrites your own wording.
Each helper is gated behind a global switch (AI_PROVIDER,
CONTENT_TRANSLATION_PROVIDER), and the live word-cloud/open-text analyses
additionally require a per-question opt-in. All AI traffic goes only to the
endpoint the operator configures. See
docs/concept.md and .env.prod.example for the
configuration knobs.
Django 5 + Django REST Framework + PostgreSQL · React + Vite + TypeScript + Tailwind CSS · SSE for realtime · OIDC via Keycloak · LTI 1.3 via pylti1p3 · Docker Compose. Rationale: ADR-0001.
Developed and maintained by virtUOS — Universität Osnabrück (https://www.virtuos.uni-osnabrueck.de/).
Contact / maintainer: Rüdiger Rolf · rrolf@uni-osnabrueck.de
Questions, ideas and contributions are welcome — please open an issue or merge request in this project.
Apache License 2.0 — see LICENSE.