Tensorlake Sandbox

license

Apache-2.0(OSI)

source

public(SDK, tl CLI and client crates only

managed-tier

Tensorlake hosted cloud + proprietary Lattice scheduler

core features withheld

yes — the MicroVM runtime and the Lattice cluster scheduler that creates sandboxes exist in no public repo - the Apache-2.0 tree publishes the SDK, the tl CLI and client crates, which authenticate to cloud.tensorlake.ai with an API key

Tensorlake publishes its Python SDK and the client side of its CLI under Apache-2.0, but the Firecracker microVM runtime and the Lattice scheduler that create sandboxes are in no public repository, and the CLI's filesystem and Git crates are built from private sources. Getting started requires an API key for Tensorlake's own cloud, and none of the company's other public repositories implement the missing pieces.

• https://api.github.com/orgs/tensorlakeai/repos?per_page=100 recorded 2026-08-12

  All 58 public repos in the org. Beyond the SDK they are skills, cookbooks, examples, benchmarks, integration plugins and forks of third-party projects (cloud-hypervisor, vllm, arrow-rs, moka, fuser). The former open compute engine survives only as archived clients - indexify-typescript-client and indexify-extractor-template, both archived. Nothing implements Lattice or the sandbox runtime.

• https://api.github.com/repos/tensorlakeai/tensorlake recorded 2026-08-12

  Repo metadata: language Python, license spdx_id Apache-2.0, 986 stars, default branch main.

• https://api.github.com/repos/tensorlakeai/tensorlake/git/trees/main?recursive=1 recorded 2026-08-12

  Full untruncated recursive tree, 2,350 entries, one governing LICENSE plus five vendored third-party ones. Everything in it is client side: src/tensorlake/ is the SDK around cloud_client.py, and all nine crates/ are cli, cloud-sdk, function-agent-core, gsvc-codec, gsvc-fs-client, gsvc-mount and the node/py bindings. platform/ holds only macos/tlfs, a FUSE mount helper. No scheduler, allocator, VMM or control plane, and no ee/ or enterprise/ path.

• https://github.com/tensorlakeai/tensorlake recorded 2026-07-09

  Apache-2.0 license; sandbox runtime + SDK (Python + Rust); BYOC self-host

• https://raw.githubusercontent.com/tensorlakeai/tensorlake/main/Cargo.toml recorded 2026-09-29

  The workspace manifest lists only client crates as members and states "Resolution-only placeholder for the private gsvc-codec packfile codec" and "Private mount core. The source is injected only for official full builds." so parts of the official CLI are built from private crates; license is Apache-2.0.

• https://raw.githubusercontent.com/tensorlakeai/tensorlake/main/LICENSE recorded 2026-09-29

  The repository LICENSE file is the Apache License, "Apache License", "Version 2.0, January 2004", and carries the text "TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION"; searched for a second license or enterprise carve-out in the file: none.

• https://raw.githubusercontent.com/tensorlakeai/tensorlake/main/README.md recorded 2026-09-29

  The README sells the platform as server side, "Sandboxes created in under a second via Lattice, a dynamic cluster scheduler" and "Supports up to 5 million sandboxes in a single project", and setup is to sign up at the hosted cloud and "export TENSORLAKE_API_KEY" for the tl CLI; searched the body for self-host, on-prem, BYOC, enterprise and license key: none.

• https://www.tensorlake.ai/ recorded 2026-07-09

  sandbox-native cloud for AI agents; run sandboxes in your own cloud account

GitHub stars are the only usage signal published for Tensorlake's sandbox repository, since no download, install or customer figure exists. A star is not a use, so this reads as attention rather than adoption volume.

• https://api.github.com/repos/tensorlakeai/tensorlake recorded 2026-10-17

  stargazers_count = 1023 for tensorlakeai/tensorlake

Tensorlake has most of the features expected from a leading sandbox platform: Firecracker microVM isolation for every tool call, snapshots that can be paused, forked and resumed, versioned mountable volumes, and durable functions with queues, timers and retries. Independent ComputeSDK benchmarks put its median time-to-interactive at 1.13 seconds, slower than the vendor's sub-second claim, although latency is not the main differentiator in this category. Scale is the bigger gap: Tensorlake reports thousands of concurrent sandboxes, while E2B has recorded more than a billion sandbox starts.

• https://docs.tensorlake.ai/sandboxes/introduction recorded 2026-09-29

  The sandbox docs introduction says "they are MicroVMs backed by Firecracker and CloudHypervisor" and that "tensorlake/ubuntu-minimal" starts in "a few hundred milliseconds".

• https://docs.tensorlake.ai/sandboxes/snapshots recorded 2026-09-29

  The snapshot docs describe a memory checkpoint that will "restores filesystem, memory, and running processes exactly as they were" and a filesystem checkpoint that is the default.

• https://www.computesdk.com/benchmarks/sandboxes/ recorded 2026-09-29

  The ComputeSDK sandbox benchmarks index says "Independently verified sandbox benchmarks run on a recurring schedule." and lists Tensorlake in the Sandbox Burst TTI composite ranking at 88.4 dated Sep 25, 2026.

• https://www.computesdk.com/benchmarks/sandboxes/burst-tti/ recorded 2026-09-29

  The ComputeSDK burst time-to-interactive page, "A leaderboard of common benchmarks for each of our sandbox providers.", with the last run "September 25, 2026" in "Northern Virginia, US" with 4 vCPU / 16GB RAM, lists Tensorlake at 88.4 composite, "1.13s" median, "1.20s" P95 and "1.21s" P99, each at 100 percent success; searched for a 1.35s figure: none.

• https://www.tensorlake.ai/ recorded 2026-09-29

  The homepage describes "isolated Firecracker microVM sandboxes that scale from one to thousands" with snapshot, suspend/resume and durable orchestration, and lists "QUEUES, TIMERS, AND RETRIES" among the orchestration features.