Privacy Policy
This policy explains what CodeTrain collects and how we use it.
Who we are
CodeTrain is operated by InferHaven LLC, a limited liability company registered in the State of Michigan, United States, trading as InferHaven. InferHaven LLC is the data controller for the personal data described here.
Postal address: InferHaven LLC, PO Box 145, Bessemer, MI 49911, United States.
Privacy contact: [email protected]. We do not operate
a public phone line, and email is the fastest way to reach a person.
This policy is governed by the laws of the State of Michigan, United States. If you live somewhere with its own mandatory data protection law, such as the UK, the EU or California, that law still applies to you and nothing here limits the rights it gives you.
- Free skill: The free skill runs inside your own coding agent and makes no calls to CodeTrain or InferHaven servers: its page talks only to a server on your own machine (
127.0.0.1), with no account and no telemetry. Your agent sends what it reads to its own model provider, under that agent's and provider's terms; in a lesson that includes your code, the lesson file and a short summary of your local profile, and the rest of the profile remains local unless you ask the tutor to use it. The only other network requests are Pyodide and the packages of it your code imports, from the jsDelivr CDN the first time each is needed. Your code is not sent there, and nothing is fetched from PyPI. - The
codetrainagent (local): your source code stays on your machine. Only the small prompt snippets needed for a tutoring turn are sent through our model proxy to the AI provider, or on bring-your-own-key straight from your machine to the provider, never touching our servers. We don't copy or store your repository. - Managed in-browser tutor: the code you submit for review and any files you attach as context are sent to our servers and on to the AI model provider, and stored in your lesson session so you can resume it and see your history. Python and JavaScript you type run in your browser; bash and PHP run in an ephemeral, network-isolated sandbox on our infrastructure.
We don't use your code or lessons to train AI models, and we only use model providers under arrangements that don't train on your prompts.
What we collect
- Account data: via our auth provider (Clerk): your email, name, and (if you sign in with GitHub/Google) basic profile info.
- How you found us: when you first arrive through a link with a campaign tag (ref or utm_source), your browser stores that tag; with no tag, it stores only the name of the website that sent you, for example perplexity.ai, never the full address. It is sent once, when your account is created. The first source recorded is kept; later visits do not change it. We use it to count which channels bring people to us.
- Usage & metering: session counts, model token usage and cost, plan and entitlement state, so we can enforce limits and show your usage.
- Billing: handled by Stripe. We store a customer/subscription reference; we never see or store your full card number.
- Lesson content (managed in-browser tutor): the code you submit, notes and questions you write, and any files you attach as context. We send these to the AI model to generate feedback and store them in your session so you can resume it and review your history.
- Learning path data: joining a learning path is optional, and so is every answer. If you join, we store which path you joined and when, and your answers to the path's questions: why you are learning, where you run things, how you rate yourself in Linux, networking and Docker, about how much time a week you have, how you like to learn, and an optional goal in your own words. For each concept on the path we store practice counts: steps answered right without help, right with help, and missed, when you last practised it, and the tutor's estimate of how well you know it. When the tutor spots a common mistaken belief in your written answer, or you pick a concept-check option that shows one, the path notes which belief it was, where it was noted (the lesson and the step, never your words), how sure the tutor was, how many times, and whether a later answer resolved it; it also schedules a review of that concept. The memory panel on the Path tab shows all of it.
- Lab command log: a lab is a lesson where you work in a terminal on a real Linux machine of your own. While you work, the lab keeps a log of the commands you ran: at most the 100 most recent commands, each command line cut at 500 characters, each command's exit code and the times it started and finished, and the diagnosis you chose and whether it was right. The log also keeps a short part of each command's output: up to 2 KB, with colour codes removed, for the commands the lab looks for as evidence or as a change to the machine, and up to 200 characters for any other command. The raw terminal stream, every byte the terminal showed or received, is never stored, logged or sent to anyone.
- Prompt data: sent to the AI model provider to generate a tutoring response, under arrangements that do not train on your prompts.
- Support communications: if you email us.
What we do with it
To operate and secure the Service, enforce plan limits, process payments, provide support, and improve the product. We do not use your code, prompts, or lessons to train AI models, and we do not sell your personal data.
AI processing & code execution
By default your turns use Anthropic's Claude (Haiku/Sonnet) models, sent directly to Anthropic's API. If you choose one of the optional experimental models, your prompt is instead routed through our model gateway (OpenRouter) to that model's provider (for example OpenAI, Alibaba/Qwen, Z.ai, or Moonshot); the picker flags these. In every case we use arrangements that don't train on your prompts. When a lesson runs bash or PHP, your code executes in a short-lived, network-isolated container on our sandbox host and is deleted right after.
A path lesson sends the AI model your typed answers (why you are learning, where you run things, your self-ratings, how you like to learn, your time a week) so the tutor can pitch the lesson, and the beliefs the path has noted about that lesson's concepts. Your goal in your own words is never sent to the model. The same model providers and the same no-training arrangements apply as for every lesson.
Each lab runs on its own virtual machine on our sandbox host, with no network connection to anything outside it. After the lab, the lesson asks you to explain what failed (the explain-back). For that step only, the AI model is sent a summary of the lab: at most the 40 most recent commands, each cut at 200 characters, with up to 300 characters of output for the evidence commands, and the diagnosis attempts. The same model providers and the same no-training arrangements apply as for every lesson. Checking the diagnosis and checking the fix on the machine never call an AI model.
Email we send you
Some email is operational and you cannot opt out of it while you have an account: sign-in and security messages, team invitations, billing receipts, and notices about material changes to this policy or the Terms.
We may also send occasional product email to the address on your account, such as a note about a feature you have not tried or an offer. Every one of these carries an unsubscribe link and supports one-click unsubscribe in your mail client, and opting out is permanent unless you ask us to resume. Opting out of product email never affects the operational messages above or your use of the Service. We do not sell or rent your address, and we do not send you other companies' advertising. Our mailing address is PO Box 145, Bessemer, MI 49911.
Service providers
We share the minimum necessary with: Clerk (authentication), Stripe (payments), Anthropic (the default Claude models), OpenRouter (model gateway for the optional experimental models) and the providers it routes to for those (e.g. OpenAI, Google, Alibaba/Qwen, Z.ai, or Moonshot), Cloudflare (web hosting/CDN/DNS), Fly.io (API hosting), Neon (database), Oracle Cloud (the sandbox host that runs the server-side language lessons), and Sentry (error monitoring). Each processes data under its own terms.
Cookies
We use essential cookies for authentication/session management (via Clerk). We don't use advertising trackers.
Data retention
How long each category is kept. "Life of account" means until you delete the account or ask us to.
| Data | Kept for |
|---|---|
| Account record (email, name) | Life of account. |
| Lesson history and submitted code | Life of account, so you can resume a lesson and review what you did. Removed when the account is deleted. |
| Learning path data | Life of account. You can change, delete or remove parts of it before then, from the memory panel on the Path tab. Removed when the account is deleted. |
| Usage and metering | Retained after account deletion where we need it for accounting and tax records, with the personal identifiers removed. |
| Billing records | Held by Stripe under its own retention schedule. We store only a customer and subscription reference. |
| Error reports | Held by Sentry under its retention schedule. These carry operational metadata only, never prompts, model replies or your code. See how that is enforced. |
| Sandbox execution | Not retained. Bash, Ruby, Perl, PHP, Node and OpenWrt run in a container that is destroyed straight after the run. |
| Lab machines and command logs | The lab machine and its disk are destroyed when the lab step is finished, after 30 minutes without use, after 2 hours, or when its disk grows past 2 GB, whichever comes first. A machine is never kept or reused for anybody else. The lab's command log is stored with the lesson, in your lesson history: kept for the life of the account and erased when the account is deleted. |
| Database backups | Encrypted dumps taken twice daily and kept for 7 days, then deleted. Our database provider additionally holds a 6 hour point-in-time window. Deleted data can persist in a backup until it ages out of both. |
| Email suppression | If you unsubscribe we keep that record indefinitely, including after account deletion. It exists so we cannot email you again by mistake, and deleting it would undo the thing you asked for. |
Deleting your account
You can do it yourself, immediately, from Plan & billing in your account dashboard. It takes effect at once rather than entering a queue.
What is erased: your lesson history and every piece of code and writing inside it, including the lab command logs, your saved concepts and revision schedule, your learning path data (your answers, the per-concept practice counts, and the noted beliefs), your profile, feedback you submitted, your API tokens and linked devices, and the identifying fields on your account record, including your name and email address.
What survives, and why: usage totals stay, because they are our accounting records, but they are detached from any identifying information and cannot be traced back to you. A one-way hash of your email address is kept so the same address cannot be used to open a new free account, which is the control that stops the free tier being reset by deleting and re-registering. If you later want to come back, email [email protected].
Backups are not rewritten, so a copy can persist until the backup it sits in ages out on the schedule above. If you would rather we handled it, [email protected] still works.
Your rights
Depending on where you live (e.g. GDPR/UK GDPR, CCPA), you may have rights to access, correct, export, or delete your personal data, and to object to certain processing. Email us to exercise them.
The learning path data can also be seen, changed, removed and exported from the memory panel on the Path tab, without emailing us. The per-concept practice counts can be seen and exported there, not deleted one by one.
Security
We use encryption in transit, scoped access tokens, and reputable infrastructure providers. No system is perfectly secure, but we work to protect your data. The engineering detail, including which files we refuse to read and what our error monitoring is not allowed to collect, is on the security page.
Children
The Service isn't directed to children under 16, and we don't knowingly collect their data.
Changes
We'll post updates here with a new date. Material changes may also be notified by email.
Contact
Privacy questions or data requests: [email protected].