Tigris Object Storage

Agent Kit is a TypeScript library that simplifies storage workflows for AI agents by bundling @tigrisdata/storage and @tigrisdata/iam into four primary primitives:

• Forks: Copy-on-write clones of a dataset for isolated, writable copies.
• Workspaces: Per-agent buckets with optional TTL and scoped credentials.
• Checkpoints: Snapshots of bucket state for restoration into fresh forks.
• Coordination: Webhooks on bucket events to trigger multi-agent pipelines.

Tigris uses S3-compatible APIs and cannot read from Azure Blob Storage directly. To migrate, you must use s3proxy, which acts as a translation layer: it presents an S3 endpoint to Tigris and translates S3 requests into Azure Blob Storage requests.

Tigris supports lazy migration via shadow buckets, meaning it fetches objects from the s3proxy endpoint on demand and caches them, eliminating the need for upfront data copying and ensuring zero downtime. Optionally, write-through mode can be enabled to sync new writes back to the Azure Blob container during the migration process.

Tigris provides Git-style workflows for data management using Snapshots and Forks. Both are zero-copy and instant (O(1)) because they use an append-only architecture that captures references rather than duplicating data.

• Snapshots: Immutable, point-in-time records of an entire bucket. They are used for auditing, recovery, or as a seed for a fork.
• Forks: Isolated clones of a bucket created from a snapshot or the current state. Forks allow for experimentation or testing without affecting the source bucket; you only pay for the changes made to the fork, not the shared baseline data.

A fork creates a new bucket from a snapshot or the current state of an existing bucket. Unlike copying, forks share baseline data by reference (zero-copy), meaning only new or modified objects are stored.

Key characteristics:

• Instant Creation: No data is copied at creation time.
• Isolation: Writes to a fork are isolated and do not affect the source bucket.
• Divergence: Forked buckets diverge from the source at the moment of the fork.
• Billing: You only pay for new object versions written to the fork; shared baseline data incurs no extra charges.
• Dependencies: Source buckets cannot be deleted while forks depend on them; all forks must be deleted first.

Tigris Acceleration Gateway (TAG) is an open-source, S3-compatible caching proxy designed to sit between an application and Tigris object storage. It transparently caches objects on local NVMe storage using an embedded RocksDB cache, reducing read latency to sub-millisecond levels without requiring application code changes. Users simply point their S3 client to the TAG endpoint instead of the Tigris endpoint.

Key benefits include:

• Zero Code Changes: Works with any S3-compatible library, AWS SDK, or AWS CLI.
• Performance: Eliminates network round-trips for repeated reads, saturating high-bandwidth NICs (e.g., 85+ Gbps on 100 Gbps NICs).
• Transparent Proxy: Forwards SigV4 credentials to Tigris without storing secret keys.

Presigned URLs grant temporary access to private objects without requiring credentials. They are used to share private files, enable direct browser uploads, or provide time-limited download links.

Key constraints and features:

• Expiration: URLs can last up to 90 days (default is 1 hour/3600 seconds).
• Operations: Supports both GET (download) and PUT (upload) access.
• Custom Domains: Presigned URLs work with custom domains by replacing the Tigris domain in the generated URL with the custom domain.