"A new distributed storage system built for elasticity, iteration, and safety: a transactional distributed block store, lineage-aware storage proxies, and a user-space block device driver, with versioned copy-on-write and application-consistent snapshots coordinated with Postgres. Thanks to Fluid Storage, Agentic Postgres delivers fast, zero-copy forks on real production data, which makes it possible to create instant environments, to experiment safely in sandboxed environments, and to run agentic loops directly on real data."
"Agentic Postgres' MCP server allows agents and developers to interact with the database through high-level prompts such as: "I want to create a personal assistant app. Please create a free service on Tiger. Then using Postgres best practices, describe the schema you would create." Tiger Data says they have distilled more than a decade of Postgres experience into a set of built-in master prompts covering schema design, query tuning, migrations, native full-text and semantic search, and more."
"Full-text and semantic search are powered by two Postgres plugins: an improved version of the existing pgvectorscale, which delivers higher-throughput indexing, better recall, and lower latency at scale; and the new pg_textsearch, which implements BM25 for modern ranked keyword search and is optimized for hybrid AI workflows. Currently, pg_textsearch runs in-memory to ensure maximum speed, with disk-based operation support in development."
Agentic Postgres is a Postgres-based database tailored for AI agents and developers, offering a CLI for terminal access and an MCP server for high-level prompt interaction. Built-in master prompts encode over a decade of Postgres best practices for schema design, query tuning, migrations, and search. Full-text and semantic search use two plugins: an improved pgvectorscale for high-throughput vector indexing and pg_textsearch implementing BM25, currently in-memory for speed. Fluid Storage provides a transactional distributed block store, lineage-aware proxies, a user-space block device driver, versioned copy-on-write, and application-consistent snapshots coordinated with Postgres to enable fast, zero-copy forks on production data.
Read at InfoQ
Unable to calculate read time
Collection
[
|
...
]