Published on June 30, 2026, the paper proposes organizing information retrieval systems with indexes, models, and inference on users’ devices, while making remote services optional.
The paper “As We May Search” proposes a local-first approach to information retrieval: indexes, models, and inference run on users’ devices, while remote services remain optional. Its framework organizes retrieval architecture decisions around privacy and other dimensions, focusing on engineers assessing systems that handle sensitive data.
The publication states its proposal and scope, but the available material does not provide experimental results or performance measurements. To assess its claims, consult the original paper and check its criteria, examples, and evidence. If you use AI to study or apply its ideas, avoid entering sensitive organizational data; use fictional or anonymized examples and follow your internal data policy.