About Semora

Every date you need for the next four months is already written down before the term starts. That is the strange thing about a college semester. The midterm date, the final project weight, whether Friday is a lecture or a lab, what percentage the participation grade is worth — all of it is printed in a document your professor handed you in week one. The information problem is solved. The transcription problem is not.

Because that is what standing between you and an organized semester actually is: transcription. Four or five syllabi, eight to twenty pages each, and the dates you need are scattered between the attendance policy and the academic-integrity statement. Copying them into a calendar by hand takes an evening you do not have in the first week of term, so almost nobody does it. The PDFs stay in the email attachment. The dates live in your head.

The failure that follows is not forgetfulness, and it does not happen in week one. It happens in week six, when two exams land in the same 48 hours and a project you half-remembered turns out to be worth 25 percent. By then the problem is not that you did not know — it is that you never saw the shape of the term, so you could not have started earlier even if you had wanted to.

So Semora is built around deleting the transcription step and nothing else. You handle each syllabus once, in about the time it takes to photograph five pages. What comes out is not a summary or a chat window: it is structured rows (dated deadlines with weights, meeting blocks with days and rooms, letter-grade cutoffs) because rows are what a calendar, a weighted average and a workload calculation can actually be built on. Everything else the app does is downstream of that one step.

• The input is whatever you have, and it differs by surface: on iPhone and iPad, a photo of up to five pages or a PDF picked from Files; on the web app, a drag-and-dropped file or text pasted straight into the scanner, between 20 and 60,000 characters. The paste path is web-only, because that is the surface with a keyboard and a clipboard rather than a camera.
• A photo scan carries a size budget of 10 MB across all its pages, checked in the app as each page is added, so you find out one page early rather than after capturing everything.
• The output is structured: course and instructor, meeting and office-hour blocks, term start and end, the grading scale, and every dated item with a type, a weight and a confidence score.
• Nothing about your deadlines saves silently. The course, its meeting times and its grading scale are written for you; the deadlines wait on a review screen where you edit, deselect and approve them.
• Anything the model was unsure about is badged rather than hidden — items scored under 0.8 confidence, and dates that fall outside the term.
• Undated items are not thrown away and not given an invented date. They sit deselected in a group called "Needs a date."

The obvious alternative is to read Canvas. Plenty of planners do exactly that, and it is a reasonable design — the assignments are already there, already dated, already tied to a course. Semora starts somewhere else on purpose, and the reason is worth stating precisely rather than as a slogan.

A learning management system tells you what has been posted. A syllabus tells you what is coming. In week one, a Canvas course page is frequently close to empty; the term is real but the assignments have not been created yet. The syllabus, on the other hand, is complete on day one by definition, because a professor has to publish the whole term's plan before it starts. If the point is to see the shape of the semester before it happens, the syllabus is the only document that contains it.

The syllabus also carries things a course page usually does not. Percentage weights per item, which is what makes a weighted average possible at all — a grade tracker that does not know the midterm is worth 30 percent is just a list of scores. The letter-grade cutoffs your specific professor uses, which is what turns 87.4 into a B+ rather than a guess. The meeting blocks with days, times and rooms, and the office hours, including the by-appointment ones with a location and no fixed days. Term start and end dates, which is how a date that parses to next February gets flagged as suspicious instead of quietly filed.

There is a second reason, and it is structural rather than editorial. Building on someone else's platform means availability depends on that platform and the institution running it. Semora supports Canvas, Blackboard, and Moodle import as a Pro feature. Its current Canvas connector uses a personal access token and should be used only where the institution permits third-party token connections. If it is unavailable or not permitted, scan the syllabus or paste the assignment list into the web scanner. An LMS connection is a useful second input; it is a poor foundation for the product.

If you are on the free tier and your deadlines live in Canvas anyway, there is a path that costs nothing: sign in to the web app on a laptop, open your Canvas assignments page, select the text, and paste it into the scanner there. Same extraction, same review screen, one of your five monthly scans. It is the web app specifically, because pasting text is not an entry point on iPhone or iPad.

The apps are one codebase. React Native through Expo compiles to a universal iOS app that runs on iPhone and iPad, and the same source builds the web app you sign in to in a browser. That is why the surfaces do not drift apart: a fix to the grade math is a fix everywhere, not three separate implementations that disagree in the third decimal place. The surfaces differ only where the hardware differs — the camera and device calendar sync on iOS, drag-and-drop and pasted text on the web.

Data sits in Postgres on Supabase, with authentication handled by the same service. Anything that costs money or must not be trusted to the device runs server-side in an edge function: the syllabus parse, the tutor, flashcard generation, LMS import, receipt validation. The client asks; the server decides.

The extraction itself uses OpenAI GPT-5.6 Luna through the Responses API. The model is not asked for prose. It is asked for one structured JSON object with named fields, and every field is validated on the server before it reaches your account, so an unparseable date becomes a null you can fix on the review screen rather than a corrupt row inside your course. Each call is logged with its outcome, which is also how the scan quota is counted.

None of that is exotic. It is worth writing down because the architecture is the reason for two things you can actually observe: the free limits behave identically on iPhone, iPad and web, and a Pro-only feature cannot be turned on by a client that decides to skip a check.

Your account data — semesters, courses, deadlines, grades, notes — lives in a Postgres database with row-level security switched on for every personal table. In practice that means each table carries a policy comparing your authenticated user id to the row's owner, so a request for somebody else's rows does not return a permission error to work around; it returns nothing at all. That check happens in the database, underneath the API, so it holds regardless of what a client sends.

Files you upload are handled the same way. Syllabi and course notes go into private storage buckets where the path begins with your own user id, and the read policy requires that folder to match the account making the request. Syllabus documents routinely contain names, emails, office locations and schedules, which is exactly the kind of content that should never sit behind a guessable public URL.

Syllabus content and any notes you attach are sent to the OpenAI API for processing by GPT-5.6 Luna. OpenAI states that API data is not used to train its models unless a customer explicitly opts in. Semora disables response storage, while OpenAI may still retain abuse-monitoring logs for up to 30 days unless a stricter retention control applies.

Product analytics are deliberately blunt. Events are app-level (a scan completed, a paywall viewed) and they are tagged with a random identifier generated on first install. There is no user id on those rows, so usage data is not tied to your account, your name or your email. It is enough to know which screens are used and where people get stuck, and not enough to reconstruct a person.

Deleting is self-serve and permanent. Settings has a Delete Account screen that requires a hardware identity check (Face ID, Touch ID or your passcode) and a fresh sign-in before anything happens, because an irreversible action should not be reachable by anyone with thirty seconds of physical access to your phone. It then removes your uploaded files, deletes your rows, and deletes the auth user itself. Nothing is archived and nothing is recoverable afterward, which is the point.

• Row-level security on every personal table, scoped to your authenticated user id.
• Private storage buckets for syllabi and course notes, path-scoped to your account folder.
• Syllabus and note content sent to OpenAI for processing with response storage disabled; API data is not used for training unless explicitly opted in.
• Semora session tokens use the device's secure storage on iOS.
• Anonymous, device-scoped analytics with no user id attached.
• In-app account deletion behind a biometric or passcode check plus a fresh sign-in. Files, rows and the account itself, all removed.

This belongs on an About page rather than a pricing page, because it is a statement about how the product treats you rather than what it costs. The free tier's limits are not screens that hide buttons. They are database triggers.

A free account gets 5 syllabus scans per calendar month, up to 4 courses, and one semester in total. The scan limit is checked in three separate places — in the app, in the parsing function on the server before any paid extraction runs, and again by a trigger on insert. The course limit and the semester limit are checked in two: in the app, and by a BEFORE INSERT trigger that refuses the row outright. Pro removes the monthly scan cap and leaves one fair-use ceiling behind it, around 20 scans in any rolling 24 hours, which exists to stop scripted abuse of a paid endpoint and which a normal term does not approach.

This cuts in both directions, and that is why it is worth telling you. It means nobody gets Pro features by patching a client, which is the part that keeps the price at $3.99 a month. It also means the limits are exactly what this site says they are, with no quiet grace period and no soft edge, including the one that is genuinely unflattering. One semester means one, for the life of the account, not one at a time. A free account cannot start a second term in January. The only way to free that slot is to delete the finished semester, which cascades through its courses, deadlines and grades and archives nothing.

We would rather you read that sentence here than discover it in January. A limit you find out about at the moment it blocks you is a bad experience regardless of how reasonable the limit is.

Marketing pages drift from products. It happens honestly — a feature moves behind a paywall, a limit changes, and the page describing it was written six months earlier by someone reading a different page. So this site is written against the shipping source instead of against other marketing copy. When a page here says a feature is Pro, that claim was checked against the actual gate: the server-side entitlement check, the database migration, the constant in the file. When it states a number, the number came from the code that enforces it.

That method is not decoration, and the proof is that it caught real errors. On the first pass, four separate tier claims on this site were wrong, all four in the flattering direction, and all four were corrected against the app: whether Canvas import is free, whether hosting a Course Space is free, whether device calendar sync is free, and the one-semester cap on free accounts. Each of those would have been a student installing the app on a promise and hitting a paywall at the first tap.

The same standard requires disclosing limits that do not help us. That free is one semester in total. That the fifth course stops you whether it arrives by scan, by hand, or by joining a classmate's Course Space. That hosting a Course Space is Pro, while joining one is free. That device calendar sync writes to your phone's calendar and does not run in a browser, where the equivalent is downloading an .ics file, which the web app does do. That Pro is purchased inside the iOS app, so there is no web checkout: if you never use an iPhone or iPad, there is currently no way to buy it.

Comparison pages carry an extra rule, because they describe products we do not build. Every factual claim about a named competitor comes from that company's own site or store listing, is attributed as such, and is hedged where the behavior is not publicly confirmed — those rows say so in the table rather than guessing. Nothing about a competitor is inferred, and nothing is invented to make a column look worse. Where a competitor is genuinely better on a point, including on price, the page says so.

And there is a rule about what this site will never contain. No testimonials, no star ratings, no user counts, no university logos, no awards, no press quotes. Not as a stylistic choice — those things simply do not exist yet, and inventing them is the single most common way a new product lies. If you find an error anywhere on this site, email it and it gets fixed.

One subscription, two prices. Pro is $3.99 a month or $19.99 a year, which works out to about $1.67 a month. It is purchased inside the iOS app through Apple's StoreKit, and the entitlement attaches to your account rather than your device, so signing in to the web app with the same account turns Pro on there with no second charge and nothing to activate.

That is the entire business model. There are no ads anywhere in the app or on this site, and no advertising SDK in the build. Your data is not sold, brokered or shared with data partners — the third parties involved are the ones required to run the thing, and they are listed in the privacy policy: the database and auth provider, the AI processing API, Apple for purchases, and the push delivery service.

It is also worth being straightforward about why the free tier is metered the way it is. Every scan is a paid model call, and so is every tutor answer and every generated flashcard deck. Those costs are per use, not per user, which is why the free limit lands on scanning rather than on the parts that cost nothing to run. Deadlines, tasks and subtasks are unlimited on free and always have been. Grade tracking, weighted averages and semester GPA are free. Same-day reminders are free. Joining a Course Space a classmate hosts is free, permanently.

The free tier has no expiry date, either. The scan counter refills on the 1st of every calendar month, measured in UTC, and free features do not degrade if you never subscribe. If you subscribe and then stop, nothing is deleted. Your courses, deadlines, grades and notes stay readable and editable, and the free limits reapply only when you add something new. What does change is access to the Pro surfaces: flashcard decks, the Workload dashboard, Smart Plan, Progress Insights, the AI tutor and device calendar sync go back behind the paywall until you resubscribe. The data behind them is still sitting there (a lapsed subscription pauses an active calendar sync rather than unwriting it, and a deck you built is locked, not erased) but you cannot open those screens again until Pro is back on the account.

Semora is independent. It is not the product of a company you have heard of, it has no outside funding to point at, and it launched recently enough that there is no track record to point at either. Every claim on this site is about the product, because the product is the only thing there is evidence for.

There is a real upside to that, and it is the one thing worth selling here. A bug report goes to the person who wrote the code, not to a queue. Feature requests genuinely move the order of what gets built, because the roadmap is not defending a quarterly plan. The fact-checking method described above exists precisely because a small product cannot afford a reputation for overstating what it does.

The honest downsides belong in the same paragraph. There is no Android app and no Mac app; the platforms are iPhone, iPad and any browser. Google Calendar and Google Classroom sync are not shipped, and the device calendar sync that does exist is iOS-only — in a browser the equivalent is exporting an .ics file, which the web app downloads directly. Pasting syllabus text into the scanner is likewise web-only. There is no phone support, no live chat and no enterprise agreement. Response times are a person's response times.

Email is semora365@gmail.com, and it is the same address for support, bug reports, privacy questions, data export requests and corrections to this website. If you are reporting something broken, the three things that make it fixable fastest are which surface you were on (iPhone, iPad or browser), what you tapped immediately before it happened, and what you expected instead. Deleting your account does not require emailing anyone — it is in Settings, and it is immediate.