Tensorlake Sandboxes

What it does

Tensorlake Sandboxes provides on-demand, isolated microVM environments for agent harnesses, tool calls, coding agents, builds, and other workloads. Developers create a sandbox through the API, Python or TypeScript SDK, or CLI, selecting an image and resources; commands run inside it and return process results such as standard output. Sandboxes can retain memory and filesystem state across suspend and resume, and users can take snapshots or clone an environment.

The product is aimed at developers and teams building or evaluating agents, running code generated by models, and operating development or CI environments. Hosted access uses Tensorlake’s cloud service and requires an account and API key; developers can work through SDKs or CLI, and named sandboxes also support SSH and remote IDE access. Additional services include versioned filesystems, hosted Git, and orchestration. Free access is limited to one concurrent sandbox, 1 vCPU, 1 GB RAM, 10 GB disk, and sessions of up to two hours. GPU sandboxes use gVisor; CPU-only sandboxes run in VMs.

What it can do

• Create isolated sandboxes on demand with a chosen image and resource configuration
• Execute shell commands and tool calls in sandboxed environments
• Suspend and resume sandboxes while preserving memory and filesystem state
• Snapshot and clone sandbox environments
• Mount versioned filesystems and hosted Git workspaces
• Connect to named sandboxes over SSH or remote IDE tools

Works with

• Harbor
• OpenCode
• Crabbox
• Claude Managed Agents
• Cursor Cloud Agents
• Devin Outposts
• Scale AI AgentEnv

Before you choose

• The free tier allows one sandbox at a time, up to 1 vCPU, 1 GB RAM, 10 GB disk, and two hours per session.
• Paid usage is metered by active CPU, allocated RAM and disk, and suspended snapshot storage; activity may be blocked when prepaid credits run out.
• GPU sandboxes use gVisor, while CPU-only sandboxes run in VMs.
• CPU charges fall back to allocated vCPU time when active CPU samples are unavailable or untrustworthy.

Listed under

• Coding
• Sandboxes and infrastructure
• Agent infrastructure
• AI agents
• Cloud compute
• Code execution
• Developers
• Rl rollouts

Get started

The official site links to account signup and API-key access; the free tier has limits.

1. Create a Tensorlake account and obtain an API key.
2. Install the Python package or TypeScript SDK, or use the CLI.
3. Create a sandbox with an image and resource settings, then run commands through the SDK or CLI.

Free tier: $0 forever. Usage Credits: prepaid $5, $10, or $20 packs, with no postpaid overage. Pro: $250 per monthly billing cycle, including 25,000 CU; additional usage is $0.01/CU. Enterprise pricing is custom. Usage meters include active CPU per core-hour, RAM per GB-hour, disk per GB-hour, and suspended snapshot storage per GB per 30-day month. Network egress is stated to be free. Active CPU runtime is metered when available; if samples are unavailable or untrustworthy, allocated vCPU time is billed instead.

Source: tensorlake.ai/pricing, checked 1 Nov 2026. Prices can change; check before you buy.

Sources for this overview

Product pages, documentation, and pricing checked 1 Nov 2026. Each source supports part of the overview; availability and prices can change.

• Tensorlake Documentation ↗docs.tensorlake.ai
• Tensorlake — Sandboxes for AI Agents ↗www.tensorlake.ai
• Pricing — Tensorlake ↗www.tensorlake.ai
• Tensorlake — Sandboxes for AI Agents ↗www.tensorlake.ai

Company website

tensorlake.ai: product, pricing and policy pages

YC directory mirror

Batch and company name

Social accounts

Accounts the company links, and its posts on X

Human review

Checked by a person

knowledge_base

manual_review

Last checked 1 Nov 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.