Loading
Cadreen is made by Billingrails Inc.
Billingrails Inc. collects and processes your data solely to run the work you give Cadreen, which includes training your workspace's own models as described below. This includes the requests your systems send, the services you connect, the rules you set, the decisions people make, and the records of what ran. We do not sell, share, or use your data for advertising. Your requests are processed by the outside companies whose AI models do the work, and by the services you connect, under their own data handling policies, which may differ from ours.
Training is part of the service, and it is on by default. The prompts your workspace sends and the answers Cadreen recorded in it may be used to train your workspace's own models, including ones from before training was on. Approvals of actions Cadreen proposed are used for training under the same switch, when they are linked to a recorded operation, from the newest 10,000 decisions recorded in each workspace. Each approval is kept as one line of what was proposed, with emails, phone numbers, card numbers and ID numbers hidden, and for an action the software asked to take only the action's name, never its details. Who answered is not part of what trains. Approvals that kept no line of what was proposed are not used. Switching training off stops them being read for later runs. Rejections are kept as your record. A preference run that an admin starts, and agrees the cost of, can also use a pair where a proposal was rejected and a revised one approved, as two short lines (each cut to 400 characters, with emails, phone numbers, card numbers and ID numbers hidden). Most pairs are made by a judge model comparing two answers, and you can look at a sample of the pairs, and see how many there are, before training is approved. Nothing trains by itself: a person has to start a training run, and for a run the data goes to the training provider used for that run. A preference run also sends cleaned questions and two answers to a judge model, from one of our model providers, to build the pairs, and that is charged to your workspace.
After a run, the Training quiz asks the new model questions written from your documents and shows how close it now is. An AI grader from one of our model providers compares the questions, your team's answers and the model's answers, before and after training; the questions and your team's answers are cleaned first, as the training data is, and an answer that still holds personal data is left out and not sent. Every version of your model that a run trains is kept, so you can go back to any of them. A version is deleted when you delete its run or the workspace.
Before a run, names, email addresses, phone numbers (7 and 10 digits), addresses and dates of birth are masked, in English text only; text that isn't clearly English, and very short examples, are left out. Company names are not masked. Credentials and secrets we recognise (keys, tokens, passwords and the like) are masked too. We may not recognise every one, so some can remain.
A workspace admin or the owner can switch training off at any time on the Training page. Off takes effect at the next export: a run that has already exported its data finishes, and nothing new is used after that. Your data trains only your own organisation's models: a model of the workspace it belongs to, or, if an admin of every workspace involved includes it in a run, a model of the workspace that run was started in. We never use it to train anyone else's model, or our own. To train your model, we send it to our training provider. Their contract with us forbids them from using it to train or improve their models or anyone else's. Like most providers, they keep usage records (such as how much was processed) and may review content for safety and legal reasons. A run may include several workspaces only when the person starting it is an admin or owner of every one of them, and only workspaces that have training on are used. Such a run trains a model of the workspace it was started in.
We retain your data only as long as necessary to deliver the service, comply with legal obligations, and resolve disputes. You may request deletion of your data at any time. Usage and billing records are retained for financial compliance. Security measures include encrypting the sign-in details and keys we store, keeping each customer's data separate from every other customer's, and recording all significant actions in a log.
Two calls to our API work without an account. The check (POST /api/v1/public/assess) uses your network address to limit how often it runs. The contact call (POST /api/v1/maas/leads) keeps the name, email, company and use case it is sent, so a person can reply. Our home page uses neither.
For any privacy concerns, data access requests, or deletion requests, contact privacy@billingrails.co.
By using Cadreen, you agree that Billingrails Inc. provides an AI-driven execution service — not professional advice, legal counsel, medical guidance, or financial recommendations. You retain full responsibility for outcomes produced through your instructions, including the rules you set and the actions you approve. We do not guarantee specific results. Your account may be suspended for abuse, fraud, or violation of applicable law.
You own your inputs and outputs. Billingrails retains a limited license to process, store, and transmit them solely to deliver the service, which includes training as described in the privacy policy above.
The work in a workspace (what it sends and records, and the decisions its team approves) can be used to train that workspace's own model. This is on by default; an admin can turn it off.
While Cadreen is in its launch period it may not charge for some or all of this. Before charging starts for your workspace, we will tell you.
While charges are off, each workspace can use a limited amount a day for free. When it is used up, work waits until 00:00 UTC.
Your Cadreen balance is in US dollars, and a new balance starts at $0. It pays for the model calls made through Cadreen, at Cadreen's prices and for what each call used, and for training runs. Calls through the gateway on keys Cadreen manages for you are charged at what the provider charged. Calls through the gateway on your own provider keys are not charged by Cadreen. Cadreen's own dashboard chat is paid by Cadreen, within a daily limit.
Before a training run's data is sent, you are shown the most the run could cost. It is charged to your balance when the run ends, for what it used and never more than that amount. A run that stops early is charged for what ran, and a run that used nothing is not charged.
Model calls and training runs are checked against your balance before they start, and refused if it can't cover them (calls through the gateway on keys Cadreen manages are charged afterwards, and can leave an amount owed). If a charge is more than your balance holds, what it holds is taken and the rest is owed: new work waits while you owe more than your balance holds, and the next money you add goes to what is owed first.
Cadreen does not refund failed work automatically. If you think a charge is wrong, email intelligence@billingrails.co.
Adding money is not taxed for now.
For questions about these terms, contact intelligence@billingrails.co.