Adapters

Tensorlake

Overview

Tensorlake Cloud Volumes are durable, versioned file trees. One Tensorlake filesystem maps to one storagesdk location.

Uploads use checksum-attested URLs, so bytes flow directly from the SDK process to blob storage and do not traverse the Tensorlake API server. ReadableStream bodies are spooled to a temporary local file first, keeping multi-GiB uploads bounded in memory. Copy, move, snapshot, and fork operations reuse immutable content references and publish metadata only.

Configuration

import { Storage } from '@storagesdk/core';
import { tensorlake } from '@storagesdk/adapters/tensorlake';

const storage = new Storage({
  adapter: tensorlake({
    filesystem: 'agent-runs',
    apiKey: process.env.TENSORLAKE_API_KEY,
    organizationId: process.env.TENSORLAKE_ORGANIZATION_ID,
    projectId: process.env.TENSORLAKE_PROJECT_ID,
  }),
});
tensorlake({
  filesystem: string;
  apiKey?: string;          // falls back to TENSORLAKE_API_KEY / TENSORLAKE_PAT
  apiUrl?: string;
  organizationId?: string;
  projectId?: string;
});

Create the filesystem with the Tensorlake CLI or FilesystemClient.create() before constructing the adapter.

Native fast paths

Notes

Compatibility

CapabilitySupport
UploadsNative direct-to-blob publication
Copy / moveNative metadata-only mutations
SnapshotsNative metadata-only retention
ForksNative shared-storage-network fork
Byte-range readsServer-side HTTP range
User metadataNot persisted
Signed URLsNon-fetchable locator only