AI AGENT STORE / SERVICES

COMPUTE & SANDBOXES · FOR AI AGENTS

Tensorlake

Run agent code in hosted sandboxes with controlled resources and resumable state.

Tensorlake supplies sandbox and orchestration infrastructure for agents. Its API, Python and TypeScript SDKs, and CLI create execution environments, run commands, move files and manage named sandboxes that can suspend and resume.

Your integration checklist

What you need from your owner
Create an account and select a project and plan. Issue a project API key for SDK use, or authenticate the CLI separately. Set resource limits, timeouts and a budget before scheduling work.
Authentication
SDKs read TENSORLAKE_API_KEY. CLI login authenticates the CLI separately; it does not automatically configure SDK credentials.
Your first task
Create a short-lived sandbox, run a small command, retrieve its output and files, then explicitly verify that the environment has stopped.
Billing & payment
A limited free tier, prepaid credits and paid plans are available. Paid usage meters CPU, memory, disk and snapshot storage. Optional auto-refill can create additional purchases; confirm plan allowances and overages in the account.
Agent payment support
Not verified. Programmatic usage does not establish autonomous checkout support.
Operational limits
Unnamed sandboxes are ephemeral and cannot suspend. Named environments support resume, subject to plan and capacity limits. Suspended state can incur storage charges. CPU and GPU sandboxes use different isolation mechanisms.

ORIGINAL AI AGENT STORE INTEGRATION PLAYBOOK

Choose the sandbox lifecycle before starting code

Editorial recommendation · Reviewed 2026-11-03

When to choose this service

Choose Tensorlake when an agent needs isolated hosted execution or resumable environments. Existing local execution may suffice for a small trusted task without persistence or scale requirements.

Provisioned compute, managed isolation and retained execution state require operating infrastructure. An agent can write the workload but cannot provide remote capacity merely by generating code.

Suggested workflow

  1. Choose ephemeral execution for a disposable task or a named sandbox for work that must resume. Set an explicit timeout and the smallest adequate resource allocation.
  2. Wait for the running state before sending commands. Persist the sandbox identifier and collect exit status, logs and output artifacts as separate results.
  3. Verify required artifacts before stopping or suspending the environment. Review any remaining snapshot storage and resume permissions after the task ends.

Acceptance checks before real work

  • A created sandbox may still be pending and unable to accept commands.
  • A successful command exit does not guarantee the requested artifact is correct or complete.
  • Suspending compute is not equivalent to deleting retained state or stopping every charge.

If the workflow fails

Inspect the existing sandbox after an ambiguous create or resume response before allocating another. On capacity or resume failure, preserve identifiers and error details; export critical artifacts rather than relying solely on a future resume.

Provider documentation

Explore services for adjacent needs

Continue building your toolkit