Canonn, Privacy Policy and Terms of Service

Applies to the Canonn API (api.canonn.ai), canonn.ai, and the Canonn dashboard.

1. Data we collect

Category	What it is, and why
Account data	Your email and authentication identity, handled by our identity provider. We never receive or store your password.
Usage metadata	Per request: timestamp, key name, model, token counts, latency, status, finish reason, request id, IP address and user agent. This powers your dashboard and your bill.
Prompts and completions	Only if you enable I/O logging, which is off by default and can be switched off at any time in the dashboard. See §3.
Payment data	Handled by our payment processor. Card details are tokenised in your browser and never reach Canonn servers; we hold a token reference, the card brand, and the last four digits.
Support correspondence	Email you send us, kept so we can answer and follow up.

2. How we use it

To operate the API: routing, authentication, and returning your answers.
To meter usage and bill you accurately.
To detect abuse, fraud and attacks against the service.
To show you your own usage and logs in the dashboard.
To answer your support requests.

We do not sell personal data, and we do not share it with advertisers. We do not perform automated decision-making that produces legal effects for you.

3. Training policy

We do not train on your prompts or your completions. Not by default, not silently, and not as a condition of any plan. Your inputs and outputs are never used to train or fine-tune Canonn models, and are never sold or shared for anyone else's training.

The only exception would be a separate written agreement that you sign, for your own benefit — for example if you asked us to build a model tuned to your documents. Absent that document, the answer is no.

Our models are trained on public datasets, on publicly accessible documentation from the open web, and on data we generate ourselves. Customer traffic plays no part in it.

Evaluation, stated precisely

Training and evaluation are different things and we describe them separately, because a blanket "never used for anything" would not be true. Your prompts and completions are not used to evaluate Canonn models. What we do measure against, and only where the customer has agreed in writing, is a knowledge base — the documents an account has loaded — so that we can test retrieval and answer quality against real material rather than benchmarks alone. When we do:

we publish aggregate scores only — never the customer's name, and never a document, a question or an answer;
the evaluation data is not added to any training set, and does not enter a model;
you can decline, and withdraw, at any time by writing to hello@canonn.ai.

I/O logging

Prompt and completion bodies are stored only while you have enabled I/O logging on your account. It exists so you can debug your own integration; it is off by default, visible as a switch in the dashboard, and disabling it stops new capture immediately. Bodies are truncated to reasonable size limits before storage.

4. Retention

Data	Kept for
Prompts and completions (if logging is on)	30 days, then deleted automatically by a scheduled job.
Usage metadata	While your account is open, then as long as accounting and tax rules require.
Account and billing records	While your account is open, then as required by law.
Support email	Up to 24 months.

5. Security and where inference runs

All traffic is encrypted in transit with TLS.
API keys are stored only as SHA-256 hashes. A key is displayed once at creation and cannot be recovered afterwards — only revoked and replaced.
Inference runs on GPU capacity in Canada (Hyperstack, CANADA-1 region). Authentication, rate limiting and metering run on Cloudflare's global edge network; the service's database runs on Cloudflare D1, primary location Western Europe.
We serve our own model on our own infrastructure. Your prompts are not forwarded to any third-party model provider.

6. Google user data

Canonn can import documents from Google Drive (Docs, Sheets, Slides and uploaded files), messages from Gmail, events from Google Calendar, and the captions of videos on your own YouTube channel, so that Canonn can answer questions from them. It works like this:

Drive uses the per-file scope (drive.file): Canonn only ever sees the files you pick in Google's own file picker. It cannot list or read anything else in your Drive.
YouTube uses the youtube.force-ssl scope solely to list your channel's uploads and download their caption tracks. Nothing is ever modified, posted or deleted on your channel.
Gmail uses the read-only scope (gmail.readonly) to read the messages that match a search you type, up to 100 at a time, and only when you run that import. Canonn never sends, labels, or deletes mail.
Calendar uses the read-only scope (calendar.readonly) to read events on your primary calendar for the window you import. Nothing is ever created or changed.
Content is fetched directly from Google's API to your device or browser, over TLS, and converted to text there. In the Canonn app that text is indexed locally on your device. In the Canonn dashboard the extracted text is stored in your Canonn workspace as a knowledge source, exactly like a file you upload by hand, and you can delete it at any time.
Google access tokens are short-lived and held only in memory on your device or browser tab. Canonn's servers never receive or store your Google credentials, and no third party holds them on our behalf. Disconnecting revokes the token at Google.
Google user data is never used to train models, never sold, never used for advertising, and never shared with anyone except as needed to provide the feature you asked for.

Canonn's use and transfer of information received from Google APIs adheres to the Google API Services User Data Policy, including the Limited Use requirements. The YouTube feature uses YouTube API Services; by using it you also agree to the YouTube Terms of Service, and Google's handling of your data is described in the Google Privacy Policy. You can revoke Canonn's access at any time from your Google account permissions.

7. Processors

We use a small number of vendors to run the service: a cloud edge and database provider (Cloudflare: global edge, database primary location Western Europe), a GPU compute provider (Hyperstack: Canada), an identity provider for sign-in, a token vault for source connections, and a payment processor. Each processes data only to provide its function, under contract.

8. Your rights

Subject to applicable law, including the GDPR, you may request access to your personal data, correction, deletion, portability, or restriction of processing, and you may object to processing. Write to hello@canonn.ai; we respond within 30 days. You may also complain to your local data protection authority.

You can delete logged prompt and completion bodies yourself at any time by disabling I/O logging and requesting deletion of what was already captured.

9. International transfers

Inference is served from Canada. The edge runs on Cloudflare's global network, so a request is first handled at the Cloudflare location nearest to the sender; the database behind it has its primary location in Western Europe. Other vendors (identity, payments) operate globally. Where personal data leaves the EEA we rely on adequacy decisions or standard contractual clauses.

10. Children

The service is not directed at children under 18 and we do not knowingly collect their data.

11. Changes

Material changes are announced by email to account holders and reflected in the date at the top of this page.

Between you (the customer) and Canonn.

A. The service

Canonn provides an OpenAI-compatible inference API serving the Canonn R1 model and a dashboard for keys and usage. You access the API with keys issued in your dashboard.

A2. Intended purpose and limitations

This section is the written specification of Canonn's intended purpose. It is referred to in Article 3(12) and Article 25(2) of Regulation (EU) 2024/1689 (the EU AI Act), and it forms part of these terms.

What Canonn is for. Canonn is a general-purpose AI system for retrieval-grounded question answering: it answers questions from documents you supply, and shows the passages it used. It is intended for informational and productivity purposes. Every answer is generated by an AI model and can be wrong.

What Canonn is not for. Canonn is not intended for, and must not be used for, any use listed in Annex III to the EU AI Act — including recruitment and employment decisions, evaluating creditworthiness of natural persons, life and health insurance risk assessment and pricing, access to essential public or private services, education admission or assessment, law enforcement, migration and border control, or the administration of justice.

Canonn's AI system is not to be changed into a high-risk AI system. Canonn does not supply the documentation, technical access or other assistance referred to in Article 25(2), and the obligation to do so does not apply to us.

If you nonetheless deploy Canonn into an Annex III context, or put your own name or trademark on it, or otherwise modify its intended purpose so that it becomes a high-risk AI system, then you become the provider of that high-risk AI system under Article 25(1) of the EU AI Act. The full set of provider obligations in Article 16 then falls on you: a risk management system, data governance, technical documentation, logging, instructions for use, human oversight, accuracy, robustness and cybersecurity, a quality management system, conformity assessment, CE marking, registration in the EU database, and post-market monitoring. Canonn is not the provider of that system and does not carry those obligations for you.

If you intend such a deployment, do not start it — write to hello@canonn.ai first. Where we agree to support it, we will do so only under a separate written high-risk addendum that sets out, as Article 25(4) requires, exactly which information and technical access we provide and on what terms. Absent that signed addendum, an Annex III deployment is a breach of these terms and we may suspend the account.

Human review. Whatever you build, a Canonn answer must not be the sole basis of a decision that produces legal effects for a person or similarly significantly affects them. Outputs are to be reviewed by a competent person before any such decision is taken.

B. Your content

You keep all rights to the data you send us and to the outputs you receive. We claim no ownership of either, and we assign to you whatever interest we might otherwise have in the outputs. You are responsible for the data you send and for how you use what comes back.

C. Your account and keys

You are responsible for keeping your API keys secret and for all usage made with them. Revoke a compromised key immediately in the dashboard; tell us if you believe your account has been accessed by someone else.

D. Acceptable use

You may not use the service to:

break the law, or help anyone else do so;
generate content that infringes intellectual property or privacy rights;
produce material that sexually exploits minors, incites violence, or harasses people;
attempt to extract, reconstruct or reverse-engineer our model weights;
use our outputs to train a competing model, or resell the API without our written approval;
evade rate limits, abuse the free allowance through multiple accounts, or attack the service.

Prohibited AI practices

Article 5 of the EU AI Act prohibits certain uses of AI outright, and prohibits them for the user of a system as much as for its maker. You may not use Canonn, in the Union or on people in the Union, to:

evaluate or classify people by their social behaviour or personal characteristics in a way that leads to detrimental or unjustified treatment — social scoring (Art. 5(1)(c));
assess or predict the risk that a person will commit a criminal offence on the basis of profiling or personality traits alone (Art. 5(1)(d));
infer the emotions of a person in the workplace or in education, except for medical or safety reasons (Art. 5(1)(f));
categorise people by biometric data to deduce race, political opinions, trade union membership, religious or philosophical beliefs, sex life or sexual orientation (Art. 5(1)(g));
deploy subliminal, manipulative or deceptive techniques, or exploit vulnerabilities of age, disability or social or economic situation, so as to materially distort behaviour and cause significant harm (Art. 5(1)(a), (b));
build untargeted facial-image scraping, real-time remote biometric identification, or the generation of non-consensual intimate imagery or child sexual abuse material (Art. 5(1)(e), (h), (ba), (bb)).

Disclosure and high-risk use

You may not deploy Canonn into any Annex III high-risk context without a signed high-risk addendum from us — see §A2.
If you publish Canonn output as text informing the public on matters of public interest, you must disclose that it is artificially generated, unless the text went through human editorial review and someone holds editorial responsibility for it (Art. 50(4)).
Where you make Canonn available to your own end users — including through a shared assistant, your own domain, or the API behind your own product — you must not remove, hide or contradict the notice that the person is interacting with an AI system. Removing "Powered by Canonn" removes our attribution only.

We may suspend an account that violates these rules, and will tell you why where we lawfully can.

E. Fees, credits and auto top-up

Usage is priced per token at the rates published in the dashboard and on canonn.ai. Prepaid credits are drawn down as you use the API.
A free monthly allowance is available without a card, so you can evaluate the model properly. Past it, credits are required.
Auto top-up, if you enable it, charges your saved card by the amount you chose when your balance falls below the threshold you set. You control both numbers and can disable it at any time.
Credits do not expire while your account is open. Except where the law requires otherwise, purchased credits are non-refundable; unused credits are not exchanged for cash.
Prices may change with 30 days' notice. Credits already purchased are honoured.
You are responsible for applicable taxes.

F. Availability

The standard plan is provided without a contractual uptime guarantee. Enterprise agreements with an SLA, dedicated capacity, and support commitments are available — contact us. We may change the API and will give reasonable notice of breaking changes.

G. Our intellectual property

We retain all rights in the Canonn service, software, model weights and brand. Nothing here transfers them to you. Access to a hosted API is a licence to use the service, not a licence to the model.

H. Warranties and disclaimer

The service is provided "as is", without warranties of merchantability, fitness for a particular purpose, or non-infringement. Model outputs can be wrong. Canonn is built to answer from the documents it is given and to decline when the answer is not there, and we publish our benchmark results including where we fall short — but it remains a tool, not professional advice. Do not rely on it as the sole basis for legal, medical, financial or safety-critical decisions.

I. Limitation of liability

To the maximum extent permitted by law, neither party is liable for indirect, incidental, special, consequential or punitive damages, or for lost profits or data. Each party's aggregate liability arising out of the service is limited to the fees you paid us in the twelve months preceding the event giving rise to the claim.

J. Indemnity

You will defend and indemnify us against third-party claims arising from your content or your use of the service in breach of these terms.

K. Confidentiality

Each party will protect the other's non-public information with reasonable care and use it only to perform under these terms.

L. Termination

You may stop using the service and delete your account at any time. We may terminate for convenience with 30 days' notice, or immediately for material breach, non-payment, or where the law requires it. On termination we delete your data in line with the retention section above.

M. Changes

We may update these terms with 30 days' notice for material changes; changes required by law take effect when required. Continuing to use the service after a change means you accept it.