Your stories don't train AIs

The problem

Almost every app you use is keeping your content to train the next generation of models. Your emails, your chats, your posts — they're fuel. When what you're keeping is the story of your life, that stops being acceptable.

What Anecdotario does

Three concrete commitments, each verifiable:

• We don't train models on your corpus. Not the in-house models we use for internal suggestions, nothing. When we ship improvements, it's not because we "read what you wrote."
• Providers don't train on your data either. When one of your anecdotes goes through an Anthropic model (for the interview or the draft) or through OpenAI's Whisper (for audio transcription), it goes under each provider's API commercial terms — which contractually forbid using inputs/outputs to train models. This has been the default since 2023, not something we opted into: it's part of the contract every API developer has.
• Local inference where possible. Operations that can run without an external LLM (full-text search, basic autocomplete, draft normalization) run in your browser or on our infra. We minimize what leaves.

The honest part

Anthropic and OpenAI retain inputs for up to 30 days by default, for abuse monitoring. That's part of the standard terms. There are "Zero Data Retention" (ZDR) plans that eliminate that window, but they require enterprise contracts we don't have today — we're an early-stage startup and the default terms cover what matters most, which is not training on your data.

If that 30-day retention bothers you, not using the app is a reasonable choice. When we have enough volume to negotiate ZDR we will, and it'll be the new default — no silent changes.

Where to verify

• Our Privacy Policy section §5 lists every sub-processor and explicitly states the no-training guarantee.
• Anthropic's commercial terms (anthropic.com/legal/commercial-terms) and OpenAI's data usage policy (platform.openai.com) are public. Anyone can read them and compare against what we claim.
• The code that calls these providers is centralized in lib/llm/client.ts with a comment restating the guarantee. If it ever changes, that file — and the Privacy Policy — must be updated in the same PR.

Why it matters

Anecdotario's value lies in being a place where you can write things you wouldn't write elsewhere. That only works if you trust that what you write doesn't become training data for some model to answer better in another context. If we can't promise that verifiably, we can't offer the app.