Freezmo: self-hosted notes-to-activities study app

Next.js 15 app that turns photos, PDFs, DOCX/PPTX and text into a swipeable
feed of AI-generated activities (flashcards, multiple choice, true/false,
fill-in-the-blank, matching, ordering, short answer) using the user's own
Ollama models.

Highlights:
- All AI calls run in server route handlers; the browser never talks to Ollama.
- Bring-your-own-key mode (default): the key lives only in the browser and is
  sent per-request; server, local-daemon and shared-key modes also supported.
- Client-side document extraction (pdf.js, mammoth, jszip) with scanned-PDF
  page rendering for the vision model; notes stay in their original language.
- ChatGPT-style shell: collapsible sidebar / mobile drawer, neutral token-based
  light + dark themes, folder organisation, PWA (manifest, SW, offline page).
- Strict JSON validation with zod + auto-repair, rate limiting, and request
  guards (cross-origin, body-size, API-key sanitising).
- setup.sh bootstraps deps, .env.local and a production build.

Gitignore excludes .env*, the generated pdf.js worker and local scratch files.
This commit is contained in:
Martin Dvořák 2026-10-06 20:56:55 +00:00
commit 3ea2cea100
83 changed files with 10909 additions and 165 deletions

226
app/api/generate/route.ts Normal file
View file

@ -0,0 +1,226 @@
import { z } from "zod";
import { extractNotes, generateActivities, notesToText, type PipelineStage } from "@/lib/ai";
import { AUTH_ERROR_MESSAGES, authErrorKind, getServerConfig, resolveAuth, resolveModelName } from "@/lib/config";
import { friendlyErrorMessage, OllamaError } from "@/lib/ollama";
import { rejectCrossOrigin, rejectOversizedBody } from "@/lib/http-guard";
import { clientKey, rateLimit, rateLimitResponse } from "@/lib/rate-limit";
import { difficultySchema } from "@/lib/schemas";
import { uid } from "@/lib/utils";
import { xpForActivity } from "@/lib/xp";
export const runtime = "nodejs";
export const dynamic = "force-dynamic";
export const maxDuration = 300;
/** ~8 MB per image after client-side resizing; up to 16 images (photos + PDF pages). */
const MAX_IMAGE_CHARS = 8_000_000;
const requestSchema = z.object({
images: z.array(z.string().min(32).max(MAX_IMAGE_CHARS)).max(16).optional(),
text: z.string().max(120_000).optional(),
difficulty: difficultySchema.default("medium"),
count: z.number().int().min(4).max(18).optional(),
focus: z.string().max(300).optional(),
/** Language the deck must be written in (usually detected by the vision step). */
language: z.string().max(60).optional(),
mode: z.enum(["create", "more"]).default("create"),
existing: z
.object({
title: z.string().max(120),
types: z.array(z.string()).max(30),
count: z.number().int().min(0).max(100),
})
.optional(),
});
type StreamEvent =
| { type: "stage"; stage: PipelineStage; message: string }
| { type: "deck"; deck: unknown; meta: Record<string, unknown> }
| { type: "error"; error: string; kind?: string };
function encoder() {
const enc = new TextEncoder();
return (event: StreamEvent) => enc.encode(`${JSON.stringify(event)}\n`);
}
/** Generous but finite: ~16 compressed images plus JSON overhead. */
const MAX_BODY_BYTES = 33554432;
export async function POST(request: Request) {
const crossOrigin = rejectCrossOrigin(request);
if (crossOrigin) return crossOrigin;
const oversized = rejectOversizedBody(request, MAX_BODY_BYTES);
if (oversized) return oversized;
const config = getServerConfig();
// Cheap requests are rejected before any work happens.
const limit = rateLimit(`generate:${clientKey(request)}`, config.rateLimits.generate, config.rateLimits.windowMs);
if (!limit.allowed) return rateLimitResponse(limit);
const auth = resolveAuth(config, request);
if ("error" in auth) {
return Response.json(
{ error: AUTH_ERROR_MESSAGES[auth.error], kind: authErrorKind(auth.error) },
{ status: 401 },
);
}
let payload: unknown;
try {
payload = await request.json();
} catch {
return Response.json({ error: "Request body was not valid JSON." }, { status: 400 });
}
const parsed = requestSchema.safeParse(payload);
if (!parsed.success) {
return Response.json(
{
error: `Invalid request: ${parsed.error.issues
.slice(0, 3)
.map((issue) => `${issue.path.join(".") || "body"} ${issue.message}`)
.join("; ")}`,
},
{ status: 400 },
);
}
const { images, text, difficulty, count, focus, mode, existing, language } = parsed.data;
const imageList = (images ?? []).map(stripDataUrlPrefix).filter(Boolean);
if (imageList.length === 0 && !text?.trim()) {
return Response.json(
{ error: "Add at least one photo of your notes or paste some text to build a deck." },
{ status: 400 },
);
}
const requestedCount = count ?? config.defaultActivityCount;
const encode = encoder();
const abort = new AbortController();
// A single hard deadline for the whole pipeline (two sequential model calls).
const globalTimer = setTimeout(() => abort.abort(), config.timeoutMs * 2);
const stream = new ReadableStream<Uint8Array>({
async start(controller) {
const send = (event: StreamEvent) => {
try {
controller.enqueue(encode(event));
} catch {
/* client went away */
}
};
try {
let sourceText: string;
let deckTitle: string | undefined;
let subject: string | undefined;
let summary: string | undefined;
let detectedLanguage: string | undefined = language;
let repaired = false;
if (imageList.length > 0) {
send({ type: "stage", stage: "vision", message: "Reading your notes…" });
const extraction = await extractNotes({
model: config.visionModel,
repairModel: config.textModel,
auth: auth.auth,
images: imageList,
text,
signal: abort.signal,
onProgress: (message) => send({ type: "stage", stage: "vision", message }),
});
sourceText = notesToText(extraction.notes);
deckTitle = extraction.notes.title;
subject = extraction.notes.subject;
summary = extraction.notes.summary;
detectedLanguage = extraction.notes.language || language;
repaired = repaired || extraction.repaired || extraction.languageRepaired;
send({
type: "stage",
stage: "generation",
message: "Finding the key ideas…",
});
} else {
sourceText = text!.trim();
send({ type: "stage", stage: "generation", message: "Finding the key ideas…" });
}
const generated = await generateActivities({
model: config.textModel,
auth: auth.auth,
notes: sourceText,
difficulty,
count: requestedCount,
language: detectedLanguage,
existing: mode === "more" ? existing : undefined,
focus,
signal: abort.signal,
});
const activities = generated.deck.activities.map((activity) => ({
...activity,
id: uid("act"),
xp: xpForActivity(activity.type, difficulty),
}));
send({
type: "deck",
deck: {
title: generated.deck.title || deckTitle || "Your study deck",
subject: generated.deck.subject ?? subject,
summary: generated.deck.summary ?? summary,
language: generated.deck.language || detectedLanguage,
sourceText,
difficulty,
activities,
},
meta: {
requested: generated.requested,
generated: activities.length,
repaired,
language: generated.deck.language || detectedLanguage || null,
visionModel: imageList.length ? resolveModelName(config.visionModel, auth.auth.directCloud) : null,
textModel: resolveModelName(config.textModel, auth.auth.directCloud),
},
});
} catch (error) {
const message = friendlyErrorMessage(error, auth.auth.baseUrl);
// Never log the request body: it may contain the user's notes.
console.error("[api/generate]", error instanceof OllamaError ? `${error.kind}: ${error.message}` : error);
send({
type: "error",
error: message,
kind: error instanceof OllamaError ? error.kind : "unknown",
});
} finally {
clearTimeout(globalTimer);
try {
controller.close();
} catch {
/* already closed */
}
}
},
cancel() {
clearTimeout(globalTimer);
abort.abort();
},
});
return new Response(stream, {
headers: {
"Content-Type": "application/x-ndjson; charset=utf-8",
"Cache-Control": "no-store, no-transform",
Connection: "keep-alive",
},
});
}
function stripDataUrlPrefix(value: string): string {
const commaIndex = value.indexOf(",");
if (value.startsWith("data:") && commaIndex !== -1) return value.slice(commaIndex + 1);
return value;
}

71
app/api/grade/route.ts Normal file
View file

@ -0,0 +1,71 @@
import { z } from "zod";
import { gradeShortAnswer } from "@/lib/ai";
import { AUTH_ERROR_MESSAGES, authErrorKind, getServerConfig, resolveAuth } from "@/lib/config";
import { friendlyErrorMessage, OllamaError } from "@/lib/ollama";
import { rejectCrossOrigin, rejectOversizedBody } from "@/lib/http-guard";
import { clientKey, rateLimit, rateLimitResponse } from "@/lib/rate-limit";
export const runtime = "nodejs";
export const dynamic = "force-dynamic";
const requestSchema = z.object({
question: z.string().min(1).max(1000),
referenceAnswer: z.string().min(1).max(2000),
keyPoints: z.array(z.string().max(300)).max(8).default([]),
userAnswer: z.string().min(1).max(2000),
/** Language of the deck so feedback comes back in the same language. */
language: z.string().max(60).optional(),
});
/** Generous but finite: ~16 compressed images plus JSON overhead. */
const MAX_BODY_BYTES = 65536;
export async function POST(request: Request) {
const crossOrigin = rejectCrossOrigin(request);
if (crossOrigin) return crossOrigin;
const oversized = rejectOversizedBody(request, MAX_BODY_BYTES);
if (oversized) return oversized;
const config = getServerConfig();
const limit = rateLimit(`grade:${clientKey(request)}`, config.rateLimits.grade, config.rateLimits.windowMs);
if (!limit.allowed) return rateLimitResponse(limit);
const auth = resolveAuth(config, request);
if ("error" in auth) {
return Response.json(
{ error: AUTH_ERROR_MESSAGES[auth.error], kind: authErrorKind(auth.error) },
{ status: 401 },
);
}
let payload: unknown;
try {
payload = await request.json();
} catch {
return Response.json({ error: "Request body was not valid JSON." }, { status: 400 });
}
const parsed = requestSchema.safeParse(payload);
if (!parsed.success) {
return Response.json({ error: "Invalid grading request." }, { status: 400 });
}
try {
const result = await gradeShortAnswer({
model: config.textModel,
auth: auth.auth,
...parsed.data,
// Grading is quick; don't let a hung request block the feed for minutes.
signal: AbortSignal.timeout(Math.min(config.timeoutMs, 120_000)),
});
return Response.json(result);
} catch (error) {
console.error("[api/grade]", error instanceof OllamaError ? `${error.kind}: ${error.message}` : error);
return Response.json(
{ error: friendlyErrorMessage(error, auth.auth.baseUrl), kind: error instanceof OllamaError ? error.kind : "unknown" },
{ status: 502 },
);
}
}

131
app/api/health/route.ts Normal file
View file

@ -0,0 +1,131 @@
import { AUTH_ERROR_MESSAGES, authErrorKind, canonicalModelName, getServerConfig, resolveAuth, resolveModelName, type AuthMode } from "@/lib/config";
import { friendlyErrorMessage, ollamaListModels, ollamaVerifyKey, OllamaError } from "@/lib/ollama";
import { clientKey, rateLimit, rateLimitResponse } from "@/lib/rate-limit";
export const runtime = "nodejs";
export const dynamic = "force-dynamic";
export type ModelHealth = {
configured: string;
resolved: string;
available: boolean | null;
note?: string;
};
export type HealthResponse = {
ok: boolean;
reachable: boolean;
mode: "local-daemon" | "direct-cloud";
authMode: AuthMode;
/** True when the browser has to provide its own key and hasn't. */
needsKey: boolean;
/** null = not checked (no key), true/false = Ollama accepted/rejected the key. */
keyAccepted: boolean | null;
baseUrl: string;
hasApiKey: boolean;
vision: ModelHealth;
text: ModelHealth;
latencyMs: number | null;
modelCount: number | null;
error: string | null;
kind?: string;
checkedAt: string;
};
export async function GET(request: Request) {
const config = getServerConfig();
const limit = rateLimit(`health:${clientKey(request)}`, config.rateLimits.health, config.rateLimits.windowMs);
if (!limit.allowed) return rateLimitResponse(limit);
const started = Date.now();
const resolvedAuth = resolveAuth(config, request);
const health: HealthResponse = {
ok: false,
reachable: false,
mode: config.directCloud ? "direct-cloud" : "local-daemon",
authMode: config.authMode,
needsKey: "error" in resolvedAuth,
keyAccepted: null,
baseUrl: config.baseUrl,
hasApiKey: !("error" in resolvedAuth) && Boolean(resolvedAuth.auth.apiKey),
vision: {
configured: config.visionModel,
resolved: resolveModelName(config.visionModel, config.directCloud),
available: null,
},
text: {
configured: config.textModel,
resolved: resolveModelName(config.textModel, config.directCloud),
available: null,
},
latencyMs: null,
modelCount: null,
error: null,
checkedAt: new Date().toISOString(),
};
if ("error" in resolvedAuth) {
health.error =
resolvedAuth.error === "missing_key"
? "No Ollama API key yet. Add your key in the settings dialog — it is stored only in this browser."
: AUTH_ERROR_MESSAGES[resolvedAuth.error];
health.kind = authErrorKind(resolvedAuth.error);
return Response.json(health satisfies HealthResponse, { headers: { "Cache-Control": "no-store" } });
}
try {
const models = await ollamaListModels(resolvedAuth.auth);
health.reachable = true;
health.latencyMs = Date.now() - started;
health.modelCount = models.length;
// A key is only "accepted" once Ollama authenticated a real request —
// model listings are public, so an actual (tiny, cached) chat proves it.
if (resolvedAuth.auth.apiKey) {
const probe = await ollamaVerifyKey(resolvedAuth.auth, config.textModel, config.healthTimeoutMs * 3);
health.keyAccepted = probe.valid;
if (probe.valid === false) {
health.error = "Ollama rejected this API key. Create a new one at ollama.com/settings/keys and paste it again.";
health.kind = "auth";
return Response.json(health satisfies HealthResponse, { headers: { "Cache-Control": "no-store" } });
}
if (probe.reason === "no_credits") {
health.error = `The key works, but ${config.textModel} is not included in this account's plan. Add usage credits or switch to a free-tier model in .env.local.`;
health.kind = "no_credits";
}
}
const names = new Set(models.map((model) => canonicalModelName(model.name)));
for (const key of ["vision", "text"] as const) {
const entry = health[key];
const wanted = canonicalModelName(entry.resolved);
const listed = names.has(wanted);
entry.available = listed;
if (!listed) {
entry.note =
health.mode === "local-daemon"
? `Not in the daemon's model list. Run \`ollama pull ${entry.resolved}\` or check your sign-in.`
: `Not returned by ${config.baseUrl}/api/tags. Check the model name and your key's plan.`;
}
}
health.ok = Boolean(health.vision.available && health.text.available && health.keyAccepted !== false);
if (!health.ok && !health.error) {
health.error = "Ollama is reachable, but at least one configured model is not available.";
}
} catch (error) {
health.error = friendlyErrorMessage(error, config.baseUrl);
health.kind = error instanceof OllamaError ? error.kind : "unknown";
if (error instanceof OllamaError && (error.kind === "auth" || error.kind === "http")) {
health.keyAccepted = false;
}
health.latencyMs = null;
}
return Response.json(health satisfies HealthResponse, {
status: 200,
headers: { "Cache-Control": "no-store" },
});
}