Perplexity details CobbleDB database built with AI coding agents
CobbleDB replaced DynamoDB, cutting median batch-read latency from 31.4 ms to 5.60 ms and p99 latency from 123 ms to 24.2 ms.
- Two engineers and hundreds of proactive AI agents built the core infrastructure in two months.
- The system partitions page keys and reads them in parallel using RocksDB MultiGet while querying alternate replicas if a response lags.
- Document state is separated across Pillar for storage in YTsaurus, Lorry for S3 update batches, and CobbleDB for serving.
- Perplexity estimates at least 20% cost savings compared to DynamoDB.
- Perplexity plans to open-source CobbleDB for teams building AI search applications.
Developers building AI retrieval systems have a reference architecture showing how tailored storage engines and autonomous coding agents can cut database latency and cloud costs.
