Answers from the documentation; checking which model is on duty
Hi - I answer from the OpenSmartRoute documentation: routing, the API, plans and quotas, self-hosting. Ask away, or open a support ticket if you need a person.
Grounded in the docs - follow a source before acting on it.
Terms of service
These terms apply to the hosted platform - the web app and API at the address in the footer - and to the browser extension and desktop app when they talk to it. The open-source software itself is licensed separately under Apache-2.0 and is not subject to them.
Last updated
A new version (2026-09-21) takes effect on Oct 30, 2026.
The terms now describe the platform as it is: a plan change is billed at once and prorated, a cancellation runs to the end of the period, pay-as-you-go credits do not expire, you may bring your own model providers, the browser extension and desktop app send counts and never your text, the enterprise availability target is the one measured on the SLA page, and you can delete your account or export your data yourself.
Until then version 2026-09-06 applies. Signed-in users are told by e-mail and in the dashboard, and asked to accept the new version once it is in effect.
1. The service
The platform routes each request you send to a model, agent, skill, persona, tool or person, returns the decision with its trace and, when you ask, executes the chosen target and returns its output. It also provides OpenAI- and Anthropic-compatible endpoints, a Model Context Protocol server, a marketplace of routing building blocks, a dashboard, a browser extension and a desktop app. What each endpoint does is documented in the platform guide and the API reference; the documentation is part of these terms where it describes limits and behaviour.
2. Accounts, sign-in and API keys
You need an account to call metered endpoints. Signing up creates a personal workspace on the free plan. Creating an account means accepting these terms and the privacy notice: the sign-up form asks for it, and people who arrive through a sign-in provider, an invitation or the command line are asked when they first open the dashboard. The version you accepted and the time are recorded and shown under Account.
You sign in with an e-mail address and password or through GitHub, Google, Microsoft, GitLab, LinkedIn or your organization's own identity provider. Two-factor authentication is available to password accounts. An account whose address is never confirmed is removed after 14 days.
API keys act as the workspace with the rights you give them. Keep them secret; you are responsible for requests made with your keys until you revoke them. Keys are shown once and cannot be recovered.
Organization workspaces have owners, admins and members. Owners are responsible for who they invite, for the organization's use of the service and for the settings they pin on their members' extension and desktop app.
One person, one account. Automated or bulk sign-ups to obtain additional free quota are not permitted.
3. Plans, quotas and payment
Each plan sets requests per calendar month (UTC) and per minute, the number of keys, tenants and seats, how long your activity is kept and which features a key may call. The current values are on the pricing page and in GET /api/v1/info. On plans without an overage rate, requests over the monthly quota are refused with 429, not billed; on plans with one, requests beyond the quota are invoiced at that rate on the next invoice while the subscription is live.
Paid plans are billed monthly or yearly in advance through Stripe; overage, where it applies, is billed in arrears. Prices are in US dollars and exclude tax: VAT, GST or sales tax due for your billing address is calculated at checkout, shown before you pay and itemised on the invoice; a tax id you enter appears on it. A free trial, where a plan offers one, is granted once per workspace without a payment method and ends by returning the workspace to the free plan unless a subscription was started before; a trial a plan states as requiring a card converts to the paid plan at its end unless cancelled before.
You can change or cancel the plan at any time under Plan and billing. Moving between paid plans - up or down - takes effect at once and the difference is prorated on the next invoice. Cancelling keeps the plan until the end of the paid period; the remainder is not refunded unless the law where you live requires it.
Plan prices and quotas may change. Existing paid subscriptions keep their price until the next renewal after we have given at least 30 days' notice on the pricing page and in the changelog. The free plan is provided as-is and may be adjusted or withdrawn with notice in the changelog.
4. Pay as you go and credits
Pay as you go has no monthly fee and no monthly quota. You buy credits in advance through Stripe, within the amounts shown at checkout, and each request is charged at the per-model price shown in the catalogue and on your activity page. A request that would spend more than your balance is refused with 402 until you top up; an automatic top-up from a saved card is optional.
Credits do not expire while the workspace exists. Every credit and charge is a line in the wallet ledger you can read under Plan and billing. Credits are not transferable between workspaces and are not redeemable for cash; a refund of unused credits follows the law where you live and is handled through support.
A workspace with a subscription cancels it before switching to pay as you go; the two are not combined.
5. Your requests and content
You keep all rights to the requests you send and the outputs you receive. You grant the platform the right to process request text to make the routing decision and, when you ask for execution, to forward it to the provider behind the chosen target. Providers act under their own terms.
By default the platform keeps metadata about each request - when, which target, tokens, cost, result - and not its text. Text is kept only where you ask for it: in conversations you continue with memory switched on, in facts you tell the assistant to remember, in a request handed to a person in your workspace's Human queue, and in the request log when your workspace policy switches payload logging on. The privacy notice lists each case, for how long it is kept and how to remove it. You must have the right to send whatever you send, including any personal data in it, and to configure the tenants and constraints that decide where it may go.
6. Your own model providers
You may run catalogue targets on your own OpenAI, Anthropic, Azure OpenAI, Bedrock, Vertex AI or self-hosted endpoints under Model providers. Requests routed to such a target are sent to your endpoint under your contract with that provider, and their cost is yours; the platform meters the request but does not resell the model. The keys you enter are encrypted at rest and used only to make those calls. You are responsible for keeping your provider agreements and for revoking a key you no longer want the platform to hold.
7. The browser extension and the desktop app
The apps are provided under these terms and under the store or licence they are installed from. They check your writing on your device; nothing leaves it until you ask for an answer, a quote or a routed reply, which are requests under section 5.
The apps report activity to your workspace - counts, token estimates, the app or site category, the target that answered - never the text. The workspace admin sees these figures on the Apps page; the person using the app can see and switch what is sent in its settings.
The apps update themselves. Pointing the desktop app at an on-premises router of your own needs a workspace on the Pro plan or higher; an address pinned by your organization's managed settings is exempt.
8. Acceptable use
You may not use the platform to:
break the law, or produce content that is illegal where you or the recipient are;
attack the service or its providers - probing, load testing without agreement, circumventing quotas, guards or plan limits, or sharing keys to pool quota;
send requests designed to extract other users' data, model weights or provider credentials;
publish marketplace listings that are misleading, infringe someone else's rights, or whose manifests point at endpoints intended to harm the people who install them;
resell the hosted service as such. Building your own product on top of it, including a paid one, is fine.
The router applies prompt-injection and personal-data guards to every request. They are a safety measure for the people whose requests are routed, not a substitute for your own review of what your application sends.
9. Marketplace
By publishing a listing you grant every user the right to download and use its manifest under the licence you declare on the listing. You warrant that you have the right to grant it.
Published manifests are frozen per version so that installs are reproducible; you can archive a listing, but existing installs keep the versions they downloaded.
Paid listings are bought through the platform's Stripe checkout and installed once the payment is confirmed. How a publisher is paid for a paid listing is agreed with the publisher in writing before the listing goes live; nothing is owed for a listing removed for breaching these terms.
Reviews must be about the listing. Listings and reviews that breach these terms are removed; the publisher is told why and may resubmit after fixing the problem.
Installing a listing downloads a document; nothing runs on your behalf. Review what a template or agent points at before applying it.
10. Availability, support and changes
The hosted platform is provided without an availability commitment on the free, pay-as-you-go, pro and team plans. The enterprise plan carries a monthly availability target and a support response time, both stated on the pricing page and measured for your workspace at GET /api/v1/sla and under Plan and billing; service credits, where agreed, follow the enterprise agreement. Support response targets for the other plans are on the pricing page and are targets, not commitments.
Endpoints in the API reference are stable within a major version; when one is removed it is deprecated first in the changelog with at least 90 days' notice. GET /healthz, the status page and the changelog are the places to check when something looks wrong.
11. Suspension and termination
You can stop using the service at any time; deleting the account and a workspace is described on the privacy page. We may suspend a workspace that breaches section 8, that has unpaid invoices after a reminder, or that endangers the service or other users, and will say why. Where the breach can be fixed you get the chance to fix it first. On termination your keys stop working and the apps sign out; your data is deleted as described on the privacy page.
12. Open source
The software behind the platform is OpenSmartRoute, licensed under Apache-2.0 and published on PyPI. Nothing in these terms restricts your rights under that licence. Self-hosting the same software - including the enterprise edition - is not covered by these terms.
13. Warranty and liability
The hosted service and the apps are provided "as is". Routing decisions are recommendations produced by statistical models and the savings the platform reports are estimates; you remain responsible for what your application does with a decision and for the outputs of the providers you route to. To the extent the law allows, the operators of the platform are not liable for indirect or consequential loss, and total liability for any claim is limited to the fees you paid for the service in the twelve months before the claim. Nothing here limits liability that cannot be limited by law.
14. Changes to these terms
These terms are part of the platform and every version is listed below with the date it takes effect. Material changes are announced at least 30 days before they apply: by e-mail to every account holder, as a notification in the dashboard and in the changelog. Once a new version is in effect the dashboard asks you to review and accept it before you continue; API keys keep working in the meantime so that running systems are not interrupted. If you do not agree with a change you can stop using the service and delete your account as described on the privacy page. Questions go to support@opensmartroute.ai or the support page.
Version history
Each version is identified by the date of the change. Your acceptance of a version is recorded when you sign up or when the dashboard asks you to review a change; you can see it under Account in the dashboard.
Version
Effective
Status
What changed
2026-10-01
Nov 1, 2026
superseded
One clause changed, about free trials: every paid plan now starts with a free trial that needs no payment method, granted once per workspace; when it ends the workspace returns to the free plan unless you subscribed before - nothing converts to a paid plan on its own. Only a trial a plan expressly states as requiring a card converts at its end unless cancelled.
2026-09-21
Oct 30, 2026
announced
The terms now describe the platform as it is: a plan change is billed at once and prorated, a cancellation runs to the end of the period, pay-as-you-go credits do not expire, you may bring your own model providers, the browser extension and desktop app send counts and never your text, the enterprise availability target is the one measured on the SLA page, and you can delete your account or export your data yourself.
