legal
Privacy during the beta pilot
Last updated July 2026
Beta pilot terms — pending legal review; not a substitute for counsel.
Short version: we log your prompts and the model's responses — to bill you accurately, to debug, and to make the router smarter. Your prompts are never sent to OpenAI, Anthropic, Google, or any other model vendor: the models are third-party open weights that run on our own rented GPUs. If logging isn't acceptable for a particular piece of data, don't send it.
01What we collect
| data | why |
|---|---|
| Account: your GitHub email and name | identifies your account, credits, and keys; it's the join key for everything else |
| API key records: label, creation date, first 12 characters | so you can tell your keys apart and revoke them. Full key values are never stored — shown once, then only a hash lives in the gateway |
| Request logs: prompt text, response text, model chosen, token counts, cost, timestamp, which key was used | billing, debugging, and routing quality (see 02) |
| Attachments you send: image and document content inside a request | travels and is logged the same way as prompt text |
| Billing records: amount, date, Stripe session ID | credit ledger and receipts. Card numbers stay with Stripe |
| Operational logs: IP-level request metadata at our hosts | abuse prevention and debugging; short-lived, held by Vercel and Modal |
02Why we keep prompts and responses
Three reasons, all real:
- Billing. Token counts per request are what your balance is drawn against, and we need to be able to show our work when you dispute a charge.
- Debugging. When a request fails or a model returns garbage, the prompt is the evidence.
- Improving routing. Our router decides which model should answer. We review logged prompts, label them, and use them to train and evaluate the classifier that makes that decision. So yes — your prompt text may become training data for our router. It is not used to train the language models themselves (we don't train those; they're open weights we host).
03Who else touches it
We don't sell your data, run ads, or share it with anyone for their own purposes. It is handled by the vendors that run our infrastructure:
| vendor | role |
|---|---|
| Modal | GPU + gateway hosting — where prompts are processed and logged |
| Neon (Postgres) | database: accounts, credits, API key records, request logs |
| Vercel | hosts this website and its server routes |
| GitHub | sign-in (we receive your email and name, nothing else) |
| Stripe | card payments — they handle card data, we never see it |
| Slack (optional) | ops alerts: service health only, no customer data |
We may also disclose data if the law actually requires it. Infrastructure is in the United States. We'll add vendors as the service grows and will update this list when we do.
04Cookies and tracking
One session cookie when you sign in — that's what keeps you signed in. The chat demo remembers your invite code in your browser's local storage so you don't retype it. No analytics scripts, no ad trackers, no third-party pixels, so there's nothing to opt out of and no consent banner to click.
05How long we keep it
For the duration of the pilot. Request logs stay while they're useful for billing questions and router work; account and billing records stay as long as the account exists. When the pilot ends we delete or anonymize pilot request logs, keeping only what we need for accounting. Ask us to delete your account sooner and we will — including your request logs — except for records we're required to keep (payment records) and prompts already anonymized into a router training set, which can't be traced back and unpicked.
06Your choices
- Ask what we hold on you, or ask for a copy — email us and we'll send it.
- Ask us to delete your account and logs. Same address, and we'll confirm when it's done.
- Revoke API keys yourself in the console at any time.
- The strongest control is upstream: don't send us the data in the first place. Regulated data is out of bounds — see the terms.
07Security, honestly
Traffic is TLS-encrypted, data sits in managed hosts, keys are stored hashed, and access is limited to the people running the pilot. There is no SOC 2 report, no penetration test, and no formal compliance program — this is a pilot run by a small team. Judge it accordingly. Found a hole? Please tell us at hello@itsgoodenough.ai before telling anyone else.
08Children
The service isn't for anyone under 16, and we don't knowingly collect their data.
09Changes and contact
If we change how we use your data we'll update this page and email account holders about anything material. Questions or requests: hello@itsgoodenough.ai.
Beta pilot terms — pending legal review; not a substitute for counsel. See also the terms.