← Back to all sparks
T

Tigris

DEVOPS
Velocity6.3

Tigris positions object storage as the substrate for AI agents, with forks and snapshots as the hook

object-storages3-compatibleai-agentsforks-snapshotsdata-protectiondeveloper-tools
Current state
The Tigris feed is a technical blog that mixes genuine feature launches with engineering essays and demos. Real product releases in this window — soft delete, streaming-tar bundles, expanded lifecycle rules — sit alongside deep-dive posts (objgit, Kefka, agent-shell, LangGraph agent evaluation) that showcase Tigris's fork and snapshot primitives rather than announce shipped features.
Where it's heading
Tigris is bending an S3-compatible object store toward AI-agent workloads: per-tenant bucket forks, copy-on-write disposable environments, and snapshotting recur across both its releases and its demos. The through-line is making storage cheap to fork and roll back so each agent or tenant gets an isolated, reversible workspace — with a provider-agnostic SDK aiming to carry that model beyond Tigris itself.
Prediction
Expect Tigris to keep hardening data-protection primitives (soft delete, lifecycle, snapshots) and to lean further into agent-oriented tooling built on bucket forks; the provider-agnostic SDK is the move to watch for reach beyond its own store.

Recent moves

  1. 3d ago

    Every Tenant Has a Past: Evaluating LangGraph Agents

    An engineering essay on forking and replaying LangGraph agent state to A/B-test prompt changes before shipping; a use-case demo of Tigris's fork primitive rather than a product release.

    View source ↗
  2. 10d ago

    I taught a bucket to speak git

    A deep-dive on objgit, a single-binary git server that stores repositories directly in Tigris with no disk or database; a technical showcase of what the store can back, not a shipped Tigris feature.

    View source ↗
  3. 22d ago

    Tar saved Unix backups in 1979. Now it saves your dataloader.

    Tigris bundles let clients pull thousands of objects in a single HTTP request as a streaming tar archive, cutting the one-GET-per-object overhead — a real throughput feature aimed squarely at ML dataloader workloads.

    View source ↗
  4. 24d ago

    Introducing Soft Delete for Tigris Buckets and Objects

    Soft delete for buckets and objects makes every deletion recoverable for up to 90 days once enabled — a standard but important data-protection primitive that broadens Tigris's durability story.

    View source ↗
  5. 1mo ago

    Introducing storagesdk.dev

    ⚡ SPARK

    StorageSDK launches as a provider-agnostic Node.js storage API spanning S3, R2, Azure, GCS, and Tigris with snapshots and forks built in — extending Tigris's signature fork/snapshot model across competing backends through a single interface.

    View source ↗
  6. 1mo ago

    Give your agents disposable environments in Go

    A walkthrough of Kefka, a Go userspace shell sandbox giving each AI agent a copy-on-write bucket fork plus WebAssembly tooling; an agent-environment demo built on Tigris forks, not a Tigris release.

    View source ↗