mirror of
https://github.com/lux-apps/papers.git
synced 2026-10-30 18:15:21 +00:00
zeekay
abf99d05fc
Restructured from flat .tex/.pdf files into per-paper subdirectories. Added Antje Worring as co-author on all papers. Synced PDFs to pdfs/ for download site.
44 lines
2.0 KiB
TeX
44 lines
2.0 KiB
TeX
\section{Offline-First Advantages}
|
|
\label{sec:offline}
|
|
|
|
The capsule architecture is offline-first by construction, not as an
|
|
afterthought. This produces several advantages beyond raw latency.
|
|
|
|
\subsection{Mobile and Edge Deployments}
|
|
|
|
Mobile applications frequently operate in degraded network conditions:
|
|
subway tunnels, airplane mode, rural areas with intermittent connectivity.
|
|
Firebase's offline mode handles this with a cache, but the cache is a
|
|
subset of the data, and complex queries may fail. The capsule model has
|
|
no degradation: the full database is on the device, and all operations
|
|
work identically regardless of network state.
|
|
|
|
\subsection{Data Residency}
|
|
|
|
For applications subject to data residency regulations (GDPR, HIPAA,
|
|
financial regulations), the capsule model simplifies compliance: the
|
|
data resides on the user's device. There is no question of which data
|
|
center holds the data, which jurisdiction's courts can compel access,
|
|
or whether cross-border data transfer provisions apply. The user's
|
|
device is the primary storage location.
|
|
|
|
\begin{proposition}[Data Residency by Construction]
|
|
In the capsule model, the user's data resides on the user's device by
|
|
default. Storage providers hold encrypted blobs for durability, but
|
|
the plaintext data never leaves the user's device (or TEE). This
|
|
satisfies data residency requirements without geographic provisioning
|
|
of cloud infrastructure, because the ``residence'' is the user's
|
|
device, which is by definition in the user's jurisdiction.
|
|
\end{proposition}
|
|
|
|
\subsection{Latency-Sensitive Applications}
|
|
|
|
Applications that require sub-millisecond read latency (real-time games,
|
|
audio/video processing, local AI inference with context retrieval) cannot
|
|
tolerate network round-trips. The capsule model enables these applications
|
|
by making the database local: a retrieval-augmented generation (RAG)
|
|
pipeline that queries a local vector index over a local SQLite database
|
|
operates with microsecond latency, compared to millisecond latency for
|
|
a cloud-hosted vector database.
|
|
|