# Perplexity — CobbleDB: Lower-Latency, Lower-Cost AI Search Storage

- Company: Perplexity (perplexity.ai)
- Announced: 2026-09-14
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.perplexity.ai/hub/blog/cobbledb
- Record: https://forck.live/items/10711-cobbledb-lower-latency-lower-cost-ai-search-storage
- Subject: Perplexity

Perplexity rebuilt its search storage around CobbleDB, with Pillar for durable document state and Lorry for batched updates. It reports roughly 5× lower batch-read latency in production measurements; its internal cost model estimates at least 20% savings relative to DynamoDB.

## Evidence

Verbatim from https://www.perplexity.ai/hub/blog/cobbledb:

> With help from a new internal coding agent swarm, we rebuilt this layer around CobbleDB, a key-value hot store, supported by Pillar for durable document state and publishing, and Lorry for batched updates. The redesign separates document processing from hot-store ingestion and lets us tune serving to our workload. Production before-and-after measurements show roughly 5× lower batch-read latency, and our internal cost model estimates at least 20% savings relative to DynamoDB.

---

Record: https://forck.live/items/10711-cobbledb-lower-latency-lower-cost-ai-search-storage
Catalogue: https://forck.live/llms.txt
Current issue: https://forck.live/feed.md