How to set up OpenRouter for lecture notes
A cloud model can write more detailed notes than a compact on-device one, but it means transcript text leaves your Mac and a card is attached to the account. This guide sets up OpenRouter so the spending is capped, the data is yours, and the setup is one key, not a separate account and card for every model you want to try.
What OpenRouter is, and why bother
OpenRouter is a single account and a single API key that reach almost every current cloud model, hundreds of them, through one connection. Instead of signing up with OpenAI, then Anthropic, then whoever ships next, you top up one balance and switch models from a menu.
For lecture notes, that matters for four concrete reasons:
- You can shop for a faithful model. Note-writing quality is not the same as general intelligence, and the best pick changes often. One account lets you test several models on your own lectures for pennies instead of opening accounts you will abandon.
- You get the detail ceiling without the memory cost. A large cloud model does not occupy several gigabytes of your Mac's RAM the way a big local model does. Your Mac stays free for everything else while the note is written elsewhere.
- You cap spending and lock the model in one place. A guardrail on your key sets a hard spending limit and can restrict the key to the single model you chose, so an accidental run on an expensive model cannot happen and cannot exceed your ceiling.
- You give your card to one company, not five. Fewer providers holding your payment details is simply less to track and less to leak.
This is the same tradeoff covered in how to choose an AI model for lecture notes: on-device models stay private and free, and remain the right default for most students. Reach for OpenRouter when you specifically want a higher detail ceiling than a compact local model without surrendering your Mac's memory to a large one.
What it costs, in plain numbers
OpenRouter runs on prepaid credit. You add a few dollars up front, and each lecture note draws down a fraction of that balance: anywhere from well under a cent to a few cents, depending on the model. In LectureSync's own cloud test, a full note ranged from free to roughly fifteen cents, and at least one strong result cost under a penny.
The smallest card top-up is $5 (a small processing fee applies), which at those rates lasts most students a full semester. You are buying credit, not a subscription. There is nothing to cancel, and nothing renews on its own once you follow the guardrails below.
Set it up in six steps
The goal is a setup that cannot surprise you: a small balance, no silent recharge, and one key locked down by a guardrail that caps spending, keeps your data private, and limits the key to the model you picked. Do these once, in order.
1. Create an account and add a small balance
Sign up at openrouter.ai like any website, then open the Credits page and add the minimum you are comfortable with. $5 is plenty to start. Think of this balance as the true ceiling: the account can never spend money it does not have.
2. Turn off Auto Top-Up
OpenRouter can automatically recharge your card whenever the balance runs low. That convenience quietly defeats the point of a prepaid balance, because the ceiling refills itself. In your billing settings, make sure Auto Top-Up is off. Now the balance only goes down, and you decide every time it goes back up.
3. Create an API key
Open the keys page and create a key. Give it a name you will recognize later (for example, LectureSync),
so that if it ever leaks you can disable or delete just this one without disturbing anything else. The key is a
long string that starts with sk-or-. Treat it like a password.
On its own, a key can spend your whole balance on any model. The next step is what makes it safe to hand to an app.
4. Attach a guardrail: spending, privacy, and one model
A guardrail is a rule you attach to a key (OpenRouter also lets you attach one to a whole workspace). It is where the real protection lives, and it holds three things worth setting for lecture notes:
- A spending limit. Set a hard cap with a daily, weekly, or monthly reset window. A request that would exceed it simply fails instead of charging you, so a runaway job or a leaked key cannot drain your balance. The monthly reset doubles as a budget.
- Zero Data Retention (ZDR). Turn this on and the guardrail routes only to providers that do not store your prompts and responses, and that therefore cannot train on them. It is the single switch that keeps your transcript from becoming someone's training data. You can also enable ZDR account-wide in privacy settings; the guardrail lets you enforce it on this specific key.
- A model restriction. Add an allowlist containing only the model you chose. Now the key physically cannot reach any other model, which closes off both accidental spend on an expensive flagship and any surprise about where your text goes.
With the guardrail attached, the worst case is bounded before the key ever reaches the app: it can only spend up to your cap, only touch one model, and only route to providers that keep nothing.
5. Paste the key into LectureSync
In the app, go to Settings → Connections, add OpenRouter, and paste your key. LectureSync stores it in your Mac's Keychain and sends it only to OpenRouter, never to a LectureSync server, because there isn't one. Use Test Connection to confirm the key works, then pick a model from the built-in picker.
6. Test on two or three real lectures
Do not judge a model on one clean result. Generate notes for a couple of your harder lectures (the ones with names, exact quantities, or a worked example) and check that the model kept them and did not invent anything. Faithfulness comes before polish; a confident note with a wrong calculation is worse than a plain one. The model-selection guide has a full checklist.
Your guardrail, in one place
The guardrail attached to your key is the piece that actually protects you. Set these and the worst case is bounded before the key ever leaves OpenRouter:
- Spending limit: a hard cap that fails requests over the limit; give it a monthly reset to use it as a budget.
- Zero Data Retention: route only to providers that keep nothing and cannot train on your transcript.
- Model allowlist: restrict the key to the one model you picked, so it can't wander onto an expensive one.
- Small prepaid balance: keep only what you plan to use; the account cannot spend what isn't there.
- Auto Top-Up off: so the balance never silently refills.
- One key per app: named, so a leak means disabling one key, not rebuilding your setup.
Note: the spending limit lives on the key or its guardrail, not on the account as a whole, so the guardrail is the setting that does the protecting.
Where your transcript goes
It is worth being exact about the data path, because privacy is the main thing you trade for a cloud model:
- Your Mac records and transcribes the lecture. With on-device transcription, the audio never leaves the machine.
- LectureSync sends the transcript text, plus your key, straight to OpenRouter. It does not operate a backend, so none of this passes through LectureSync.
- OpenRouter routes the request only to a provider your guardrail allows, with Zero Data Retention on, gets the note back, and returns it to the app. A ZDR provider keeps nothing after the note is written.
LectureSync never receives your recording, transcript, notes, or key. The exact data paths are listed in the LectureSync privacy policy, and the deeper version of this tradeoff is in how to transcribe lectures privately on a Mac.
Choosing a model without overspending
OpenRouter's catalog is large and changes weekly, which is a feature and a trap. Two habits keep it useful:
- Start from the tested picks, not the priciest logo. In LectureSync's lecture-specific test, price did not track note quality. An inexpensive model matched flagships on the measured task. The current recommended models page keeps the names, costs, and methodology up to date.
- Use a free variant to try before you pay. Many models offer a free, rate-limited version (its name ends in
:free). It's slower and capped in requests, but it's enough to sanity-check a model on one lecture before you spend anything.
The best model is not the most expensive one. It is the cheapest one whose notes you can still trust on a hard lecture.
If something doesn't work
| What you see | Usual cause | Fix |
|---|---|---|
| Invalid key / 401 | Key mistyped, has extra spaces, or was disabled | Re-copy the whole sk-or- key with no trailing space; confirm it is still enabled on the keys page. |
| Insufficient credits / 402 | Balance is empty, or the guardrail's spending limit was hit | Add a little credit, or raise the guardrail's limit. If it resets monthly, it will clear on its own. |
| No providers / model unavailable | Your guardrail's model allowlist or ZDR setting ruled out every available endpoint, or the model slug is wrong | Confirm the model you want is on the guardrail's allowlist; if ZDR is the blocker, pick a model with a Zero Data Retention provider (see the recommended-models page). |
| Rate limited | You're on a :free variant, which caps requests |
Switch to the paid version of the same model for real use; free variants are for testing. |
If you connect OpenRouter through a generic OpenAI-compatible option instead of the built-in OpenRouter choice,
the base URL is https://openrouter.ai/api/v1 and the key goes in an Authorization: Bearer
header. The built-in OpenRouter connection handles this for you.
FAQ
What is OpenRouter and why use it for lecture notes?
OpenRouter is one account and one API key that reach almost every current cloud AI model. For lecture notes it lets you try many note-writing models to find a faithful one, run flagship-quality models without using your Mac's memory, and lock a key to one model with a capped spend and Zero Data Retention using a guardrail, instead of handing your card to several providers separately.
How much does it cost to use OpenRouter for lecture notes?
It runs on prepaid credit, and a single lecture note usually costs from a fraction of a cent to a few cents depending on the model. The smallest card top-up is $5, which lasts most students a full semester. Some models also offer free, rate-limited variants for testing.
Can OpenRouter or its providers train on my lecture transcripts?
You control this with Zero Data Retention (ZDR). Turn ZDR on in a guardrail attached to your key, or account-wide in privacy settings, and requests route only to providers that do not store your prompts and responses and cannot train on them. OpenRouter itself does not store your prompts and responses unless you turn logging on yourself.
What is the best way to limit spending on OpenRouter?
Attach a guardrail to your key with a hard spending limit, and give it a monthly reset if you want a recurring budget. Requests over the limit fail instead of charging you. Back it with a small prepaid balance and Auto Top-Up left off, and restrict the key to one model, so a leak or a runaway job cannot drain the account.
Is it safe to put my OpenRouter key into LectureSync?
LectureSync stores the key in your Mac's Keychain and sends it only to OpenRouter. It does not run its own server, so your key, recordings, transcripts, and notes never pass through LectureSync. A guardrail on the key, capping spend and limiting it to one model, also caps the damage if the key is ever exposed.
Do I still need OpenRouter if LectureSync can run models on my Mac?
No. On-device models are the private, no-cost default and are enough for most students. OpenRouter is worth it when you want the detail ceiling of a large cloud model without giving that model several gigabytes of your Mac's memory, or when you want to compare many models cheaply.
Which OpenRouter model should I choose for notes?
Start from LectureSync's current tested cloud picks rather than the most expensive flagship, because in lecture-specific testing price did not track note quality. Run your chosen model on two or three of your own lectures and keep it only if the notes stay faithful and detailed.
One key. Every model. Notes in Obsidian.
LectureSync records the lecture, transcribes it on your Mac, and writes plain Markdown into your Obsidian vault, using an on-device model or any cloud model you connect through OpenRouter. $9.99 on the Mac App Store. Requires macOS 26 and an Apple Silicon Mac.
Get it on the Mac App Store