Is my data used to train AI?

The short answer: no.

None of your data is used to train AI models. This applies to every tool, every tier, every language — including the Free tier.

How we ensure this

Self-hosted models (Hetzner Ollama)

The default qwen2.5 + bge-m3 models run on our own GPU cluster in Helsinki. Inputs leave your browser, get processed by the model, the output is returned, and the request memory is cleared. Nothing is logged with content — only token counts and timestamps for billing.

Third-party models (OpenAI, Anthropic, xAI, Cerebras, Cloudflare)

When you use premium models, we route through their API. Each provider is contractually bound to not train on API inputs:

• OpenAI: opted out via API zero-data-retention agreement.
• Anthropic: API Terms § 4.4 — no training without explicit consent.
• xAI / Grok: API Terms § 6 — input not used for model improvement.
• Cerebras: zero-retention API tier.
• Cloudflare Workers AI: zero-retention by default.

What about embeddings?

We generate embeddings of public content (your published blog posts, the global tool catalog) for search and recommendation. These don't contain your private inputs.

If you opt in (at /account/privacy/) to use AI search across YOUR private content, we generate private embeddings — these are scoped to your account and not used for training.

What about telemetry?

We collect:

• Tool name (e.g. "blog-post-generator")
• Token count
• Latency
• Success/error code
• Anonymized hash of the model used

We don't log your inputs or outputs. Sampling for quality eval is done only on synthetic test inputs, never on user content.

Auditing

Our sub-processor list and DPA are publicly available at /legal/. Pro+ users can request a SOC 2 Type II equivalent from sales for enterprise procurement.