Parseable unifies agent, LLM, and infrastructure observability in a data lake on globally distributed Tigris object storage, with 80-90% Parquet compression.

Tigris

Parseable is an observability platform that treats logs, metrics, and AI telemetry as data lake data, storing everything in Tigris globally distributed object storage. Built in Rust, it ingests data via HTTP or Kafka, converts it to Apache Arrow in memory, then compresses it to Parquet with 80-90% compression ratios before uploading to Tigris. Queries run via Apache DataFusion with PostgreSQL-compatible SQL, using a data manifest to fetch only relevant byte ranges from object storage. A smart NVMe cache keeps recent files local to avoid round trips. The platform unifies agent behavior, LLM token costs, prompts, and infrastructure telemetry in a single queryable interface, eliminating the need for separate tools per data type. Tigris's global replication means Parseable's compute can run near query origins without managing cross-region sync.

How Parseable Built an Observability Data Lake for AI

Why AI observability needs a data lake ​