Last updated: October 19, 2026
These terms govern a sovereign Canadian AI API built for government, defence, health and regulated teams, including organizations under the Controlled Goods Program. It provides PIPEDA-aligned Canadian data residency and exact routes to the latest frontier and open-weight models.
These Terms of Service ("Terms") govern your use of the Northern Inference platform, API, and related services ("Service") operated by Northern Inference Inc. ("we", "us", "our"). By creating an account or using the Service, you agree to these Terms.
Northern Inference is an LLM API routing gateway. We route your API requests to third-party model providers (such as AWS Bedrock, Azure OpenAI, Google Vertex AI, Anthropic, and OpenAI) based on the model route you select. We are not the model provider. We provide the routing infrastructure, billing, and privacy controls.
Credit-Based Model: You purchase credits in advance. Credits are denominated in Canadian Dollars (CAD). Credits are a balance against which per-request charges are deducted at the rate in effect at the time of each request; they are not a prepayment for a fixed quantity of tokens.
Chat agent in the portal. Messages sent to the chat agent are API requests under these Terms and are charged to your account's balance at the rate shown for the model you choose. If you save chats in your browser, your browser may remove them, and we can't restore them. A monthly spending limit can be set for each person's chat. When a person reaches it, chat stops for that person until the next calendar month (UTC) or until the limit is raised. Adding credits doesn't raise it.
Saved chats and your password. Chats you save are encrypted so that only your password or your recovery code can open them. We can't read your saved chats and we can't restore them for you. Changing your password keeps your saved chats: they open with your new password straight away. If you forget your password, the reset link gives you a new password, but your saved chats stay locked until you enter your earlier password or your recovery code. Once you do, they open with your new password from then on. If you have neither, those chats are lost for good and you start again with none. Keep your recovery code somewhere safe.
Which models chat offers. The chat agent offers models in Tier 3 (processed in Canada) and Tier 4 (processed outside Canada). The model list shows each model's tier before you choose it, and each answer shows its tier and the jurisdictions recorded for it. Northern Inference can limit, at the request of your team's owner, which models a person may use and how much. Chat offers a model only if our records show that its provider has publicly committed not to use content sent to it to train models. Those commitments are the providers' own, not ours, and our Privacy Policy names the documents they rest on.
Instructions, projects, memories and templates. Personal instructions, projects with their files, memories and prompt templates are your content under section 7. When you use one in a chat, it is sent to the model with your messages and is charged as input, like any other part of a message. With memory auto-save off, a memory is saved only after you review and confirm it. If you turn auto-save on, each message you send is checked for facts worth remembering, and those facts are either saved for you or shown to you to confirm, depending on your setting. Saved memories are encrypted on your device with your password before they are stored, so we can never read them. You can review or delete any memory at any time in the Memory panel. These documents are sealed in the same way as your saved chats, so the password and recovery code rules above apply to them as well.
Ratings, export and import. You can rate an answer Helpful or Not helpful. We store the rating with your account and a reference to that answer, and not the text of the conversation. You can export your saved chats to a file. The file is not encrypted: it is yours to keep private, and we cannot recover or protect it for you. Importing a file adds its chats to your saved chats as new copies.
Images in chat (where available). Where the chat agent offers image generation, it uses a Tier 4 model provided by Google through Vertex AI. Your prompt and any images you attach are sent to Google and may be processed outside Canada. Images are charged to your balance under section 3 at the rate shown for the model. A generated image passes through our service to your browser and is saved only as part of a chat you save, sealed, in the place you chose. We do not otherwise keep it, except in the cases our Privacy Policy describes, such as Content Debug Mode. You are responsible for the images you generate and how you use them. As between you and us, you own them to the extent the law and the provider's terms allow, but we do not promise that an image is original or free of third-party rights. Generated images may carry provenance marks added by the provider; do not remove or hide them to mislead anyone.
Answers, other people's information and teams. Answers come from AI models and can be wrong, out of date or unsuitable. Check them before you rely on them. Do not put another person's personal information into chat unless you have the right to, and take particular care before you send it to a Tier 4 model. If you manage a team, you are responsible for telling its members how chat works for them, including that Content Debug Mode, if you approve it, captures their chat messages as well as yours.
Web search and web pages in chat. The chat agent can search the web and read web pages for you. One setting covers both, Web search in Chat privacy, and it is off until you turn it on. While it is off, the first time a model wants to search or read a page in an answer, the chat asks you: you can turn it on, allow it for that answer only, or decline. When you allow it, the model you chose may write short search queries, and Northern Inference sends each query to our web search provider in the United States, which returns a list of public web pages. Our Privacy Policy names the provider. The model may also ask Northern Inference to read a public web page whose address appears in the conversation or in those search results. Northern Inference fetches the page from servers in Canada and gives its text to the model. The website receives the page address and can see that Northern Inference requested it, and reading a link can count as a visit, for example on a one-time sign-in link. We show you each query and each page address. Your search queries leave Canada, including in chats with Tier 3 models, which run in Canada. Each search is charged to your balance at a flat published price in Canadian dollars, shown in Chat privacy; it is not converted from U.S. dollars. Fetching a web page is free; the text of a page the model reads is charged as input tokens on that model call. Web pages are written by third parties. They can be wrong, and they can contain text written to mislead the model. Check important information against the sources shown.
Search results. Search results in the chat come from our web search provider. You may keep them in your own chat history and use them for your own purposes. You may not copy them into a separate database or index, redistribute or resell them, or use them to create, train, evaluate or benchmark any AI model. We may suspend web search for an account that breaks these rules.
Prompt Processing: Your prompts and API request content are processed in-memory for routing purposes. We do not store prompt content by default; request logging keeps metadata only (token counts, model, latency, cost). The only exceptions are opt-in features you or your team control: PII Substitution, the Audit and Compliance tier, Content Debug Mode and scheduled monitors (see below and our Privacy Policy).
Scheduled monitors. If you create a scheduled monitor, we store the prompt you provide, encrypted at rest in a form our servers can read, so it can run while you are away. This permission applies only to the scheduled prompt you save for that monitor. It does not extend to storing your other chat messages or answers in server-readable form.
Provider Forwarding: Your prompts are forwarded to the model provider you select (or the provider serving the model route you select). Each provider has its own data handling policies. See our Privacy Policy for details.
PII Substitution: When enabled (opt-in per API key or per request), personally identifiable information in your prompts is detected and replaced with realistic substitutes before forwarding to the model provider. Original values are restored in responses. This is a best-effort feature; it may not detect all PII instances. An audit log of substitutions is available to platform administrators.
Content Debug Mode: An optional support tool, off by default, that logs the full content of your prompts and responses to help us diagnose an issue. It requires a double opt-in (a Northern Inference administrator requests it and your team owner approves by email), runs for at most 7 days before turning off automatically, is encrypted at rest, can be turned off by you at any time, and the captured content is deleted within 30 days of being turned off (or immediately on request). See our Privacy Policy section 5.
Data Residency: Every request carries a chain-of-custody record identifying the entity, region, and jurisdiction to which Northern Inference routed your data at each hop, so the route is auditable rather than assumed. Northern Inference attests this record up to the point a request enters a third-party provider. Once a request enters a provider, whether inference executes in the stated region, and how the provider handles the data, is governed by that provider's own published documentation, attestations, and contractual commitments, which we cite per route in each model's data-use posture; Northern Inference does not operate inside and cannot independently verify the provider's infrastructure. The data residency of your requests depends on the model route you select. Tier 3 models route to Canadian data centres (AWS ca-central-1, Azure Canada East), enforced at the routing layer: a Tier 3 request that cannot be served in a Canadian region is blocked rather than routed elsewhere. Each request's custody record shows the jurisdiction that served it. Processing inside a provider's region is governed by that provider's terms, which we publish for each route. Tier 4 models use provider-default or non-Canadian routes and may process data outside Canada; the custody record shows the exact route. We do not warrant that the underlying cloud regions or provider policies never change. Our infrastructure runs on AWS, a US-headquartered company operating Canadian data centres.
Tool Substitution and Local Execution: Some hosted provider tools, such as server-side web search, are not available on every model route. On Canadian-residency routes we also deliberately avoid provider-run web search, because it would send your query to a search service outside Canada. When your request includes a tool we cannot run on the selected route, Northern Inference removes that tool before sending the request to the model, and may instruct the model to accomplish the task using the tools you supplied in your own request, such as a shell or file-fetch tool that runs on your own machine. Through the API, Northern Inference does not perform web searches on your behalf, and no query is sent to a third-party search service by Northern Inference or the provider. Web search in the chat agent is described above in this section, under "Web search and web pages in chat".
Disclosure for Abuse Investigation: When an upstream provider notifies us of suspected acceptable-use violations in traffic we routed on your behalf, or when we receive a lawful request from Canadian or allied law enforcement, you authorize us to disclose your identity, account metadata, request logs, response bodies, and chain-of-custody records to the requesting provider or authority for the scope of the investigation. We do this because we are contractually the account holder on the upstream side; refusing disclosure would result in our account being suspended, impacting every customer. See our Acceptable Use Policy §6 for details of this flow.
Your use of the Service is governed by our Acceptable Use Policy (AUP), which is incorporated into these Terms by reference. The AUP enumerates prohibited content and uses in detail, describes the upstream provider policies that flow through to your traffic, and sets out our enforcement rights.
At minimum, you agree not to use the Service to:
We may suspend or terminate accounts that violate these Terms or the AUP without prior notice. Upstream providers (AWS Bedrock, Azure, Vertex AI, Anthropic, OpenAI, etc.) may independently notify us of violations in your traffic; when they do, we will investigate using your request logs, identify the responsible customer, disclose that identity and relevant records to the reporting provider, and take enforcement action per the AUP.
When your request is routed to a third-party model provider, you are bound by that provider's acceptable-use policy as if it were a direct agreement between you and them. The relevant policies are enumerated and linked in our Acceptable Use Policy §5. If any upstream policy imposes a stricter restriction than this document or our AUP for a given use, the stricter restriction governs. If an upstream provider updates its policy, the updated version takes effect for your traffic as soon as it takes effect with the provider, without separate notice from us.
We reserve the right to suspend or terminate your account immediately and without prior warning upon receiving an abuse notification from an upstream provider, for the duration required to investigate and remediate. Account credit is not refundable for termination-for-cause arising from such investigations.
You retain ownership of your prompts and the responses generated by the models you access through our Service. Northern Inference does not claim any intellectual property rights over your content.
The Northern Inference name, logo, and platform software are our intellectual property and may not be used without permission.
We provide the Service on a best-effort basis. We do not guarantee any specific uptime or availability SLA. Outages at third-party model providers (AWS, Azure, Google, OpenAI, Anthropic) are beyond our control and are not Northern Inference's responsibility.
We may perform maintenance, updates, or modifications to the Service at any time. We will make reasonable efforts to notify you of planned downtime in advance.
To the maximum extent permitted by applicable law:
You agree to indemnify, defend, and hold harmless Northern Inference and its officers, directors, employees, and agents from and against any claims, losses, damages, liabilities, costs, and expenses (including reasonable legal fees) arising out of or related to:
We may update these Terms from time to time. For material changes, we will provide at least 30 days' notice via email or a prominent notice on our website. Your continued use of the Service after changes take effect constitutes acceptance of the updated Terms.
These Terms are governed by and construed in accordance with the laws of the Province of Ontario and the federal laws of Canada applicable therein. Any disputes arising under these Terms shall be subject to the exclusive jurisdiction of the courts of Ontario, Canada.
Severability. If any provision of these Terms is held to be invalid or unenforceable in any jurisdiction, the remainder of the Terms remains in full force and effect, and the unenforceable provision will be modified by a court or arbitrator to the minimum extent necessary to make it enforceable while preserving the parties' original intent. Invalidity in one jurisdiction does not affect enforceability in any other jurisdiction.
Entire agreement. These Terms, together with the Acceptable Use Policy, the Privacy Policy, and any order forms or written agreements expressly referencing these Terms, constitute the entire agreement between you and Northern Inference regarding the Service and supersede all prior or contemporaneous communications, representations, or agreements, whether oral or written. No representation, warranty, or course of dealing not contained in these Terms forms part of the agreement.
Assignment. You may not assign or transfer your rights or obligations under these Terms without our prior written consent, and any attempted assignment without such consent is void. We may assign these Terms in connection with a merger, acquisition, financing, reorganization, or sale of all or substantially all of our assets, or to an affiliate, in each case on written notice to you.
No waiver. A failure or delay by either party to enforce any provision of these Terms is not a waiver of that provision. A waiver of any breach is not a waiver of any other or subsequent breach. No waiver is effective unless in writing and signed by an authorized representative of the waiving party.
Survival. The following sections survive termination of these Terms or your account: §3 (with respect to charges accrued or refunds owed prior to termination), §4 (with respect to data already collected), §5 and §6 (with respect to past use), §7 (Intellectual Property), §9 (Limitation of Liability), §10 (Indemnification), §12 (Governing Law), and this §13.
Notice. We may give you notice required under these Terms by email to the address associated with your account, by a banner or in-app message in our customer portal or admin UI, or by a posted update to the published version of these Terms or the relevant published rates page on our website. Notice is effective on the earlier of (a) actual receipt, or (b) the date a reasonable person in your position would have seen it given the means of notice used. You agree that electronic notice satisfies any legal requirement that notice be in writing.
Independent contractors. Nothing in these Terms creates a partnership, joint venture, employment, or agency relationship between you and Northern Inference. Each party is an independent contractor.
Headings. Section headings are for convenience only and do not affect interpretation.
Language. The English-language version of these Terms is the binding version. Any translation we provide is for convenience only.
Questions about these Terms? Contact us at support at northerninference dot ca. Abuse reports: abuse at northerninference dot ca.