Most privacy commitments are written in the language of intent. A policy says a company will not do something — until, one quarter, the incentives shift and a revised policy says it now may. Nothing about your data changed. Only the sentence describing what could be done with it did.
Pickerbird takes a different position. The boundaries you set aren’t described in a document you have to trust — they’re enforced by the software itself. When you mark something private, there is no code path that lets the AI read it. Not a discouraged one. Not one gated behind a setting. There simply isn’t one.
A policy is a promise. Code is a constraint.
The distinction matters most in the moments you’re not watching. A promise depends on the promise-keeper’s continued goodwill, their ownership structure, and their legal exposure. A constraint holds regardless — it doesn’t have a mood, a board, or an acquisition.
If a boundary can be moved without your involvement, it was never really your boundary.
This is why the three tiers — private, ask-me-first, and open — are enforced at the level where access actually happens, not in a preferences screen that a future update could quietly reinterpret.
What that looks like in practice
When the agent assembles context for a response, it can only request items whose tier permits it. A private note isn’t filtered out after retrieval — it’s never eligible for retrieval in the first place. The check is upstream, and it’s the same check every time:
// eligibility is decided before retrieval, not after
function visibleTo(agent, item) {
if (item.tier === 'private') return false;
if (item.tier === 'ask') return agent.hasConsent(item);
return true; // 'open'
}There’s no override flag, no admin bypass, no telemetry exception. That function is the whole story, and it runs the same way for everyone — including us. A few consequences fall straight out of enforcing it in code:
- It can be inspected — the rule is in the open, not in a data-handling FAQ.
- It can be tested — a private item that ever reaches retrieval is a failing test, not a policy footnote.
- It can be audited — over time, by people who don’t have to take our word.
Control you can verify
We’d rather you not take our word for any of this. A guarantee you can check is worth more than a promise you have to believe. It’s early, and we’re building in the open — if this way of thinking resonates, the beta waitlist is the best place to follow along.