Privacy
What we collect, the legal basis for each use, how to exercise your rights, and how we treat research data. Written to be readable; binding where it counts.
Disagreeable AI (“we”, “us”, “the Service”) is operated by Suncost Labs at disagreeable.ai. Application servers are located in Frankfurt, Germany (Hetzner Cloud). We are the controller of personal data processed in connection with the Service for the purposes of the EU General Data Protection Regulation (“GDPR”), the UK Data Protection Act 2018 / UK GDPR, the California Consumer Privacy Act (“CCPA”) as amended by the CPRA, and other applicable privacy laws.
For privacy questions or to exercise your rights, contact contact@suncostlabs.com. We are below the threshold that would require us to formally appoint a Data Protection Officer; the privacy contact handles all data subject requests.
We minimize data collection to what we need to operate the Service and pursue our research mission. Each processing purpose has a specific legal basis under GDPR Article 6 (and Article 9 where applicable, though we do not deliberately solicit special category data).
| Category | Purpose | Legal basis |
|---|---|---|
| Account: name, email, auth identifiers (Clerk) | Provide the Service, authenticate you | Contract (Art. 6(1)(b)) |
| Chat content: messages, assistant responses, session titles | Render the Service; show you your history | Contract (Art. 6(1)(b)) |
| Billing: Stripe customer/subscription IDs (no card numbers) | Process subscriptions, comply with tax law | Contract (Art. 6(1)(b)); legal obligation (Art. 6(1)(c)) |
| Usage telemetry: tokens, model identifiers, estimated cost, timestamps | Enforce caps, prevent abuse, internal accounting | Legitimate interests (Art. 6(1)(f)) |
| Training records: chat-turn pairs, model annotations, and quality judgments | Train, evaluate, and improve our AI models (and, where we publish, contribute to scientific research under Art. 89) | Consent (Art. 6(1)(a)) for users in the EU/UK; legitimate interests (Art. 6(1)(f)) elsewhere, balanced against your rights and revocable |
| Technical: IP, user-agent, request logs (kept ≤ 30 days) | Rate limiting, security, debugging | Legitimate interests (Art. 6(1)(f)) |
| Approximate region (CF country header) | Apply correct privacy defaults (EU vs non-EU) | Legal obligation (Art. 6(1)(c)) |
We do not sell personal data, and we do not share chat content with third-party model trainers. We do not use chat content to train our own models without your explicit research consent (see §5).
The default state of your “training and improvement” consent flag depends on the region you select at signup, in line with regional law.
You can change your consent state at any time. We log every change with a timestamp; this audit trail is part of our GDPR accountability obligation under Article 7(1).
When you allow improvement use (the consent toggle in onboarding and Settings), we may include your chat-turn records and the model-generated annotations on those turns in datasets used for the following purposes only:
Each record carries provenance metadata: dataset name, schema version, consent snapshot, judge model, and label origin. This is documented in our public source code.
We will not: sell your data, share it with third-party model trainers, or release identifiable conversations publicly. We may share aggregate, anonymized, or rigorously de-identified data with academic collaborators under data use agreements.
If you withdraw consent, we exclude your records from all future exports and training runs. Models trained before your withdrawal cannot have your data un-baked from their weights, but no new training will use your data, and your records are deleted on request (see §7).
We use the following subprocessors. Where data leaves the EU/UK/EEA, we rely on Standard Contractual Clauses or equivalent transfer mechanisms.
| Subprocessor | Purpose | Location / Transfer |
|---|---|---|
| Hetzner Cloud | Application hosting | Frankfurt, Germany (no transfer) |
| Neon (Postgres) | Durable database | EU central (Frankfurt) |
| Upstash Redis | Rate limiting, ephemeral cache | EU region (configurable) |
| Clerk | Authentication | United States (SCCs) |
| Stripe | Payment processing | United States / Ireland (SCCs) |
| OpenRouter | LLM request routing | United States; zero-retention requested where supported |
| Sentry | Error monitoring (no chat content) | United States / EU (SCCs) |
| Cloudflare | DNS, country detection (CF-IPCountry header) | Global edge; transit only |
When you submit a message, the message text is sent to OpenRouter to be routed to the underlying model provider. We request zero-retention from OpenRouter where supported; underlying providers may retain copies for short abuse-monitoring periods per their terms.
Subject to your jurisdiction, you have the following rights. We respond to verified requests within 30 days (extendable to 90 days for complex requests under GDPR Art. 12(3)).
We do not subject you to decisions based solely on automated processing that produce legal or similarly significant effects on you (GDPR Art. 22). The Service produces critique text in response to your prompts; this is not a legal, medical, financial, or other consequential decision about you.
We use a small number of cookies, all functional or strictly necessary:
We do not use third-party advertising, behavioral tracking, or analytics cookies.
The Service is not directed to children under 13 (or 16 in jurisdictions that require it). We do not knowingly collect personal data from children. If you believe a child has used the Service, contact us to request deletion.
We will notify you in-product or by email of material changes at least 14 days in advance. Material changes to research consent posture take effect only on prospective acceptance — they will not retroactively re-consent you to anything you had previously declined.
For all privacy questions, requests, or complaints: contact@suncostlabs.com. For general questions: contact@suncostlabs.com.