Tensorlake Provider

Read when:

• choosing provider: tensorlake;
• configuring Tensorlake sandbox image, snapshot, sizing, or organization;
• changing internal/providers/tensorlake.

Tensorlake is a delegated run provider. Crabbox shells out to the tensorlake CLI (tensorlake sbx ...) for sandbox lifecycle and command exec. Tensorlake owns the Firecracker MicroVM and command transport; Crabbox owns local config, repo claims, slugs, timing summaries, and normalized list/status rendering.

#When To Use

Use Tensorlake when the remote sandbox should be a Tensorlake Firecracker MicroVM and command execution should happen through tensorlake sbx exec. Use AWS, Hetzner, Static SSH, or Daytona when you need Crabbox-native SSH access.

#Prerequisites

• The tensorlake CLI must be on PATH (or pointed at via

--tensorlake-cli / tensorlake.cliPath). Install it via pip install tensorlake — the wheel ships a Rust tensorlake binary at <env>/bin/tensorlake.

• A Tensorlake API key. Crabbox passes it through to the CLI via the

TENSORLAKE_API_KEY environment variable; it is never placed on argv.

#Commands

crabbox warmup --provider tensorlake --tensorlake-image <image>
crabbox run --provider tensorlake -- pnpm test
crabbox run --provider tensorlake --id blue-lobster --shell 'pnpm install && pnpm test'
crabbox status --provider tensorlake --id blue-lobster
crabbox stop --provider tensorlake blue-lobster

#Auth

export TENSORLAKE_API_KEY=tl_apiKey_...

TENSORLAKE_API_URL (or tensorlake.apiUrl) can override the default https://api.tensorlake.ai. TENSORLAKE_ORGANIZATION_ID and TENSORLAKE_PROJECT_ID select the org/project when your account spans more than one.

#Config

provider: tensorlake
target: linux
tensorlake:
  apiUrl: https://api.tensorlake.ai
  cliPath: tensorlake
  image: ""              # ubuntu-minimal default; pin a registered image otherwise
  snapshot: ""           # snapshot ID to restore from (alternative to image)
  organizationId: ""
  projectId: ""
  namespace: ""
  workdir: /workspace/crabbox  # absolute path; sync target and -w for exec
  cpus: 1.0
  memoryMB: 1024
  diskMB: 10240
  timeoutSecs: 0
  noInternet: false

Provider flags:

--tensorlake-api-url
--tensorlake-cli
--tensorlake-image
--tensorlake-snapshot
--tensorlake-organization-id
--tensorlake-project-id
--tensorlake-namespace
--tensorlake-workdir
--tensorlake-cpus
--tensorlake-memory-mb
--tensorlake-disk-mb
--tensorlake-timeout-secs
--tensorlake-no-internet

Environment overrides (CRABBOX_TENSORLAKE_* and the corresponding TENSORLAKE_* variables read by the CLI itself) take precedence over file config. The API key is sourced from CRABBOX_TENSORLAKE_API_KEY or TENSORLAKE_API_KEY.

Run-time environment forwarding uses the normal Crabbox allowlist:

crabbox run --provider tensorlake --allow-env API_TOKEN -- printenv API_TOKEN
crabbox run --provider tensorlake --env-from-profile ~/.my-live.profile --allow-env API_TOKEN -- npm test

Crabbox prints only redacted presence/length metadata. For Tensorlake it writes the allowed values to a temporary local shell profile, uploads that file into /tmp, sources it for the command, and removes both copies best-effort after the run. Values are not placed on the local tensorlake process argv.

#Lifecycle

1. warmup / run (without --id) generates a Crabbox-owned sandbox name

(crabbox-<repo-slug>-<random6>) and runs tensorlake sbx create with the configured CPU/memory/disk/image/snapshot. The Tensorlake-assigned sandbox ID is captured from stdout and used as the canonical identifier.

2. The local lease is stored as tlsbx_<sandbox-id> with a friendly slug.
3. By default run archive-syncs the working tree: git ls-files-driven

manifest → tar -czf locally → tensorlake sbx cp upload to /tmp/crabbox-sync-*.tgz → tensorlake sbx exec -- bash -lc 'tar -xzf …' into the configured workdir. Pass --no-sync to skip the archive step.

4. The user command runs via tensorlake sbx exec -w <workdir> <id> -- <cmd>,

streaming stdout/stderr back through Crabbox.

5. On release the sandbox is terminated via tensorlake sbx terminate <id>

unless --keep was set. --keep-on-failure retains newly-created failed sandboxes and prints a rerun/stop hint.

#Capabilities

• SSH: no (Tensorlake exposes tensorlake sbx ssh directly; Crabbox does not

proxy it).

• Crabbox sync: yes — gzipped tar uploaded via tensorlake sbx cp and

extracted in-sandbox.

• Provider sync: no separate Tensorlake sync command.
• Desktop/browser/code: no Crabbox VNC/code surface.
• Actions hydration: no.
• Coordinator: no.

#Gotchas

• --sync-only, --checksum, --force-sync-large, --capture-stdout, and

--download are rejected because Tensorlake doesn't expose Crabbox's rsync semantics. Use --no-sync plus an explicit --id if you've already primed the sandbox.

• --shell wraps the command as bash -lc '<joined args>' before passing it

to tensorlake sbx exec. Plain commands containing shell metacharacters (&&, |, >, etc.) or leading KEY=VALUE env assignments are also auto-wrapped.

• Forwarded environment values are written to a temporary in-sandbox profile

for the duration of the command. Avoid forwarding broad wildcard allowlists unless you trust the sandbox and command.

• tensorlake.workdir must be absolute (default /workspace/crabbox). It's

used as both the sync target and the -w working directory for exec.

• IDs accepted by --id and stop: Crabbox slugs and tlsbx_<sandbox-id>

lease IDs that have a local Crabbox claim. Sandboxes without a local claim are rejected (matches the islo/Crabbox-owned-only safety pattern).

• The tensorlake CLI authenticates from TENSORLAKE_API_KEY in env; never

pass --api-key on the command line because process listings expose argv.

Related docs:

• Provider backends