Tensorlake Documentation

Every repository connected to Tensorlake gets a persistent cache volume. Add the Tensorlake cache action after checkout to reuse dependencies, toolchains, and build outputs across workflow runs:

- uses: actions/checkout@v6
- uses: tensorlakeai/github-runners@v1

The action detects Rust, Node.js, Go, and Python projects from their lockfiles and needs no configuration. No actions/cache step is needed. See the complete language workflow examples.

What the action caches

Language	Detected from	Cached
Rust	`Cargo.lock` or `Cargo.toml`	Cargo registry and git dependencies, the rustup toolchain, and compiled dependencies in `target`
Node.js	`package-lock.json`, `pnpm-lock.yaml`, `yarn.lock`, `bun.lock`, or `package.json`	The npm, pnpm, Yarn, or Bun download cache
Go	`go.mod`	The module cache and the build cache
Python	`uv.lock`, `poetry.lock`, `requirements*.txt`, or `pyproject.toml`	The uv, pip, or Poetry cache

Put the cache step after checkout and before any build step. Turn off the cache built into setup actions, such as cache: false for actions-rust-lang/setup-rust-toolchain and actions/setup-go, and package-manager-cache: false for actions/setup-node. To cache other directories, list them in paths:

- uses: tensorlakeai/github-runners@v1
  with:
    paths: |
      .next/cache
      ~/.cache/my-tool

How caching works

Before your build, the action downloads the job’s cache from the volume and unpacks it to the runner’s local disk, so builds never read the volume directly. After a successful job, it saves the cache if your lockfiles or toolchain changed, and waits until the upload finishes. Each branch and pull request has its own cache. A job restores its own cache first; a pull request then falls back to its base branch, and every job falls back to your default branch. Jobs save only to their own cache, so a pull request that doesn’t change dependencies reuses the default branch’s cache and saves nothing. Branch and pull request caches are deleted after a week without use. Jobs share a cache when they run the same job of the same workflow; set key to separate matrix entries that build different things:

- uses: tensorlakeai/github-runners@v1
  with:
    key: ${{ matrix.target }}

Cache scope and persistence

Jobs and branches share one cache volume per project, GitHub connection, and repository. The volume is mounted read-write at TENSORLAKE_CACHE_DIR, so code in a pull request can write to it directly. Keep secrets out of cached paths.

Don’t point CARGO_TARGET_DIR or other directories with many small files at TENSORLAKE_CACHE_DIR. Builds read those files through the network, and a warm build can be slower than a cold one.

For a small cache of a few large files, you can still use the volume directly by writing into a subdirectory of TENSORLAKE_CACHE_DIR. Changes persist after a job finishes; forced termination can discard recent writes. Use GitHub artifacts for outputs you need to keep. See the action’s README for every input.

GitHub Actions

Cache dependencies and build outputs

What the action caches

How caching works

Cache scope and persistence

​What the action caches

​How caching works

​Cache scope and persistence

What the action caches

How caching works

Cache scope and persistence