Tigris Data

One of the quiet lies of self-hosted @ClickHouseDB is that your hot data lives on your hardware. Your backups and cold partitions usually live on AWS S3, and AWS charges $0.09/GB to read your own bytes back. A 500 TB restore costs tens of thousands of dollars. That's why nobody tests their backups, and the first real test happens mid-outage. It's really a simple solution.....BACKUP TO S3 and TTL tiering are stock ClickHouse features. Point them at a bucket that doesn't charge egress and the scariest line item in your DR plan drops to zero. tigrisdata.com/blog/clickhous…

Read more on the blog: tigrisdata.com/blog/presigned…

When dealing with complicated authentication schemes, sometimes the biggest weakness in your process is actually the key thing that makes a feature possible. Learn about how presigned URLs work in object storage and how they flip a weakness into a feature!

What is an agent? Is it the LLM......the harness?....the context? We think it's the state. You can burn everything down, but you will always be able to rebuild an agent from its state files. State files can either be stored in a database, or in a plain s3 bucket. Storing checkpoints agent sources in object storage is the best way to run parallel fleets of agents, all experimenting and building up knowledge at the same time. Let us know what you think an agent is: tigrisdata.com/blog/where-doe…

Take a look: github.com/storagesdk/sto…

Object storage finally branches like Git We helped ComputeSDK build storagesdk. With this universal SDK, you can fork a bucket, let an agent or experiment run wild, then keep it or throw it away. Your production data never moves and the agent knowledge base is secure.

Loving this collaboration between @prisma Compute and Tigris.

Moving storage providers usually means writing a migration script and babysitting it. Tigris does it in one CLI command. You can point it at your old provider with a read-only key, run it, walk away. If interrupted, it resumes where it left off. To be safe, turn on write-through and new objects sync back to your old bucket until you're ready to commit. See how you can migrate at your own pace: tigrisdata.com/blog/t3-migrat…

If your storage supports forking, you can branch an agent's entire history by reference. Fork prod, point a candidate prompt at it, and replay real conversations. We just shipped a LangGraph checkpoint saver that works exactly this way. pypi.org/project/langgr…

Object storage has no connection pool to size or exhaust. Checkpoint reads and writes are plain S3 requests. Most LangGraph fleets default to a database, where tenant separation is a WHERE clause. With buckets, it's an IAM boundary. A leaked key touches one tenant, not all of them. tigrisdata.com/docs/agents/ag…

Why back agent checkpoints with object storage? For one agent on one machine, SQLite or Postgres are just fine. It starts to matter when you run a fleet of agents or a multi-tenant platform, where traffic gets bursty and parallel. tigrisdata.com/blog/eval-agen…

Put the whole world in one forkable bucket instead. Fork it per run, work against the fork, promote the winner, delete the rest. @daytonaio checkpoints the machine; the bucket checkpoints the agent. An agent you can't fork is an agent you can't trust. Blog: tigrisdata.com/blog/where-doe…

Strip an agent to parts: model, harness, sandbox, state. Three are replaceable. The state IS the agent. And today it's scattered across Postgres, Pinecone, Redis, LangGraph, and Temporal. You can't checkpoint an agent that lives in five places.

"Where does the agent run?" is a solved question: a @daytonaio sandbox that boots in 90ms and dies on purpose. The question nobody asks: if the place an agent runs is built to be destroyed, where does the agent live? tigrisdata.com/blog/where-doe…

FYI - you can reliably use object storage as a key-value store. You probably already do, you just put a database in front of it. blob in the bucket, key in the DB. We say.... drop the db. Object storage is already a KV store. tigrisdata.com/blog/tigris-kv…

Object storage finally has an undo button: tigrisdata.com/blog/soft-dele… You can turn on soft delete and every delete drops into a holding state for 7–90 days. Restore inside the window, or Tigris reclaims it for you. Your coding agent or CI job can write straight to prod, and a wrong delete is no longer a one-way door. Blog: tigrisdata.com/blog/soft-dele…

Workspace files and snapshots outlive any single sandbox, move between regions, and clone instantly. Agents start fast, run isolated, and keep state when they need it, with no storage layer for developers to manage. docs.isorun.ai/sandboxes/pers…

Tigris is happy to be partnering with @isorun to bring persistent state to their agentic runtime. Isorun gives agents fast, hardware-isolated compute. We give them durable, forkable disks, so sandboxes stay disposable, and the state that matters survives.

We built a self-updating knowledge base with Tigris! It reads your feeds nightly and posts a digest to Slack. Our product gives agents durable storage they can write to unattended, where a failed run can never corrupt what's already there. The agent writes through Tigris Shell: every change stages in an in-memory buffer and only lands on flush(). If a run throws, the buffer is dropped and your bucket is left exactly as it was. tigrisdata.com/blog/self-upda…

langgraph-checkpoint-tigris stores checkpoints on object storage, so you can fork the whole agent by reference: no data copy, constant time. Fork prod, replay real threads through the change, compare to baseline, drop the fork. github.com/tigrisdata/tig…

To catch them you need to test against the agent's real state. On Postgres that means a dump-and-restore that gets slower with every checkpoint. So most teams fall back to fixtures and ship blind.

Your agent's regressions don't show up against fixtures. They show up against real customer history, the kind a hand-written test thread never has. tigrisdata.com/blog/eval-agen…