Seshat is a free, open-source flashcard and spaced-repetition app. It exists because Quizlet put flashcards I made myself behind a paywall — so instead of paying to study my own material, I built something better and gave it away.
Most flashcard apps optimize for engagement: streaks, hearts, leaderboards. Seshat optimizes for learning.
Every non-trivial product decision — recall-first card design, the spacing algorithm, the confidence prompt,
even the typography — is backed by a citation from the cognitive-science and legibility literature, not a growth
metric. See the in-app / (home) and /attributions pages, or the research/ folder, for the
receipts.
- Sets — a named collection of cards, the core abstraction (
/setsto manage them,/sets/:idfor one set's hub page: study-mode buttons plus a random-card preview, the way Quizlet's own set page works). - Recall-first study (the default, and the one we recommend) — short-answer, cloze deletion, multiple-choice, and image-occlusion card types, confidence captured before you see the answer, FSRS spaced scheduling, and optional per-set goal dates that tighten the schedule as an exam or deadline nears.
- Confidence calibration — rate your confidence on each answer and see, over time, whether that confidence is actually justified.
- Every major Quizlet-style study mode, too — jump into any of these for a set on demand, independent of
what's due:
- Flashcards — classic flip-and-self-rate.
- Test — a generated, multi-format practice test (written, true/false, multiple-choice) across a whole set, scored at the end.
- Match — a timed term/definition matching drill. Purely a speed supplement — results aren't fed into the spaced-repetition system, unlike the other modes.
- FSRS spaced scheduling — per-card, per-learner difficulty/stability modeling (via
ts-fsrs) instead of a fixed interval table, with selectable desired-retention presets (85% / 90% / 93%). - Local-first, zero-backend — no account, no server, no tracking. Your sets, cards, review history, and
settings live entirely in your browser (text in
localStorage, images in IndexedDB) and never leave your device. Settings and keyboard remaps are also mirrored to two small cookies as a recovery net (never study data). - A real legibility system, not just a font picker —
<Legible>(src/components/Legible.tsx) is the one component every card/set surface uses to apply the whole research-backed cluster at once: Settings-driven typeface, 11.5–13pt size, 1.4–1.5 line height, and a 55–75 character measure. Seshat's own UI intentionally uses a separate, fixed typeface (Fraunces/Newsreader/IBM Plex Mono) — only the material you're actually studying gets the legibility treatment. - Portable JSON everywhere — the full Seshat format round-trips every card kind; a simpler Quizlet-style
[{term, definition}](or{name, terms: [...]}) format also imports and exports per set, for interop with plain files from other tools. One icon button imports (auto-detects the format), one exports. - Full-data backup — Settings -> Backup downloads one JSON file with everything (settings, keybindings, sets, cards, FSRS scheduling, review history) and restores it by merging or replacing.
- Keyboard-first — Cmd/Ctrl+K opens a command menu to jump anywhere; footer "Keyboard shortcuts" lists every binding; navigation keys switch between Arrow keys, WASD and HJKL, and every action is remappable.
- Agent-friendly — a
?import=URL importer, a browser-only scripting API (window.seshat, see below), WebMCP tools where the browser supports them, and JSON Schemas for imports, backups and settings.
Client-only, no backend:
- Vite + React 19 + TypeScript (strict mode)
- react-router-dom for routing — RESTfully nested:
/sets,/sets/:id,/sets/:id/edit,/sets/:id/{study,flashcards,test,learn,match}, plus a global/studyacross every set - Zod — every persisted or imported shape is validated at the storage/import boundary
- ts-fsrs — the FSRS scheduling engine
- Vitest + Testing Library for tests
pnpm install
pnpm run dev # start the dev server
pnpm run build # type-check and build for production
pnpm run test # run the test suite
pnpm run lint # lint with oxlint
pnpm run ci # every gate: format, lint, types, circular, duplication, license, coverage, build, sizeOpen the browser console on the app and call window.seshat directly — listSets(), listCards(setId),
exportSet(setId), exportSetSimple(setId), importSet(json), importSimpleJson(raw, setName?), exportAll(),
importAll(json, mode) ('merge' or 'replace'; importSet and importAll are async), plus
exportAllWithMedia() / exportSetWithMedia(setId) which embed image bytes. It reads and writes the same localStorage and IndexedDB the app does, with
no server involved (a browser tab can't run an MCP server or accept incoming connections at all — there's no
listening-socket API in JS — so this is the real "browser-only, zero-backend" version of programmatic access).
Writes dispatch a seshat:external-write event, so an open tab updates live without a reload.
Other agent entry points, all documented in AGENTS.md and public/agents.txt:
?import=<url-encoded JSON>on any URL imports a set and opens it.- WebMCP tools (
list_sets,import_all,update_settings, ...) register ondocument.modelContextwhere the browser exposes it. - JSON Schemas, generated from the Zod schemas at build time:
/schema/set-import.schema.json,/schema/seshat-backup.schema.json,/schema/seshat-settings.schema.json. Discovery files/llms.txt,/llms-full.txtand/agents.txtlink them.
GNU GPLv3 — Copyright (C) 2026 David Awad. Use it, fork it, self-host it — and any distributed modifications stay free software under the same license.
- In-app:
/(what Seshat is and why, with citations),/docs(how it works, how data is stored), and/attributions(full citation list) research/— the underlying learning-science and legibility research, one file per source, with verified links and summaries