Fluid Storage

A distributed block-storage layer for forkable, ephemeral, durable Postgres: the storage substrate behind Tiger Data’s Agentic Postgres offering, powering both the Tiger Cloud free-tier and Ghost database products.

What it is

Fluid Storage is a disaggregated block-storage system built to give Postgres (and other systems sitting on a block device) the properties that EBS and S3-based storage layers don’t:

• Zero-copy forks. A full copy of a production database can be branched in seconds, paying only for the blocks that diverge.
• True elasticity. Volumes grow and shrink without cooldowns, without over-provisioning, without “optimizing” windows.
• Synchronous replication for durability at block granularity.
• Looks like a local disk to Postgres: no application changes, no Postgres patches, full filesystem compatibility.

The public architecture is three tiers: compute nodes see a conventional block device (tier 1), a storage-proxy layer manages volumes, snapshots, and copy-on-write lineage (tier 2), and a horizontally scalable distributed block store backed by local NVMe handles durable persistence (tier 3). Blocks are persisted at 4 KB granularity, with read latency on the order of a millisecond and a single volume sustaining 110,000+ IOPS and 1.4 GB/s throughput while retaining all copy-on-write guarantees. Each cluster manages tens of thousands of volumes across workloads and tenants.

My role

I co-founded the project with Samuel Gichohi: the two of us started Fluid Storage and shaped the early architecture across the storage path, proxy layer, and Kubernetes integration.

Within the project I also authored the CSI driver that exposes Fluid Storage volumes as standard Kubernetes PersistentVolumes, the integration point that lets every Tiger Cloud database (and any future non-Postgres workload) consume Fluid Storage without code changes. 510 of 585 commits (87%) on main, over 2 years.

Public references

Tiger Data has published the public architecture and design rationale across three engineering posts:

• Fluid Storage: Forkable, Ephemeral, and Durable Infrastructure for the Age of Agents, Mike Freedman & Samuel Gichohi, Oct 2025
• Postgres for Agents, Ajay Kulkarni & Mike Freedman, Oct 2025
• Lessons from Replit and Tiger Data on Storage for Agentic Experimentation, Mike Freedman, Dec 2025

Status

Fluid Storage runs every database in Tiger Cloud’s free tier and every Ghost database today, and is the declared substrate for Tiger’s Agentic Postgres product line.

Links

Timescale/TigerData repository, described here; source not publicly linkable.