tensorlakeai/tensorlake

Tensorlake – sandbox‑native cloud for AI agents

What it is – Tensorlake is a hosted compute platform that lets you run isolated micro‑VM sandboxes (called sandboxes) at massive scale. It is aimed at developers building agentic AI applications that need a safe, reproducible environment for running LLM‑generated code, tools, or long‑running functions.

Key pieces

• Sandbox API – creates stateful MicroVMs with near‑SSD I/O, sub‑second startup, snapshot/clone, auto‑suspend, live‑migration and the ability to run up to 5 million sandboxes per project.
• Python SDK (tensorlake) and a stand‑alone CLI (tl) for creating, connecting to, copying files into, executing commands in, and terminating sandboxes. The SDK also supports pools (pre‑warmed containers) and snapshots for instant restore.
• Cloud Volumes – a versioned, object‑store‑backed filesystem that can be written to/read from via the Python, TypeScript or Rust SDKs. Files are uploaded directly to signed URLs, so payloads bypass the main API.
• Orchestration layer – serverless function runtime with fan‑out. Functions are defined with @function decorators, packaged into images, and deployed with tl deploy. Each function runs in its own sandbox and can be invoked via HTTP or the SDK.

Typical workflow

1. Sign up at cloud.tensorlake.ai and obtain an API key.
2. Install the SDK (pip install tensorlake) or the CLI (curl … | sh).
3. Create a sandbox (CLI: tl sbx create or SDK: SandboxClient.create_and_connect).
4. Run commands / copy files inside the sandbox, optionally snapshot it.
5. Deploy an application – write Python functions decorated with @function, bundle dependencies with Image, then tl deploy <path>.
6. Invoke the deployed app via tl or an HTTP request; results can be streamed with SSE.

Why it matters – Agentic AI systems often need to execute arbitrary code safely. Tensorlake provides a managed, highly‑scalable sandbox that isolates each agent, enforces network policies, and lets you snapshot state for reproducibility. The built‑in orchestration adds a serverless‑style API surface without you having to manage Kubernetes or VM fleets.

Where to learn more – Docs at https://docs.tensorlake.ai, the Slack community, and the public PyPI package tensorlake.