NeuronScope
RecipeStablesince v0.6.0

Cloud Storage Streaming (S3, GCS, Azure)

Stream, range-read, and write .nsz telemetry artifacts directly against Amazon S3, Google Cloud Storage, and Azure Blob Storage.

Audience: Cloud & Ops EngineersRead: 6 minEdit on GitHub

NeuronScope artifacts can be written to or read directly from cloud object stores (AWS S3, Google Cloud Storage, Azure Blob Storage) without writing intermediate files to local disk.

URI Schemes & Providers

bash
pip install "neuronscope[s3]"
ns save session.nsz s3://enterprise-telemetry/runs/2026/run-104.nsz

Credential Resolution

Credentials are automatically resolved via standard provider chains (AWS_PROFILE, GOOGLE_APPLICATION_CREDENTIALS, DefaultAzureCredential).

Range Reads & Zero-Copy Streaming

The .nsz container footer maps column byte ranges, allowing remote readers to fetch individual layer activations or tool call traces via HTTP Range headers without pulling the entire file.

Throughput Optimization

  • Enable concurrent chunk uploading via NS_IO_CONCURRENCY=8.
  • Place storage buckets in the same cloud region as compute clusters to minimize cross-region network latency.
Related