Globally Distributed Object Storage with Tigris

Source: fly.io Date: 2025-02-12 URL: https://fly.io/blog/tigris-public-beta/

Summary

Release announcement for Tigris, a globally distributed object storage system integrated into Fly.io. Objects under ~128KB are “instantly global” via redundant FoundationDB clusters with NVMe caching and demand-based regional replication. The API is S3-compatible, provisioned via fly storage create, and billed through Fly. The post frames Tigris as the right abstraction for a problem Fly chose not to build themselves — an explicit endorsement of the Unix philosophy applied to cloud infrastructure.

Implications

Edge deployment economics / competition with Vercel/Render. Tigris directly addresses a real developer pain: single-region S3 creates latency for globally distributed users, and building multi-region replication is complex. Fly offering a one-command solution that integrates with existing S3 SDKs is a meaningful DX improvement. For the agent workload thread, distributed object storage is the backing store for Litestream’s S3-based SQLite replication — Tigris as the default backing store would give Sprites instant global database durability out of the box. This is Fly assembling a complete agent-hosting stack: VMs (Machines/Sprites) + networking (WireGuard) + storage (Tigris + Litestream) + secrets (Macaroons).