Tuist

Tuist lets Once share cache entries across machines. Configure it when developers and coding agents should reuse the same action results.

Configure The Cache#

Add a Tuist provider and cache binding to the repository root once.toml:

toml

[infrastructures.tuist]
kind = "tuist"
account = "acme"
project = "app"

[infrastructure.cache]
provider = "tuist"

Replace acme and app with the Tuist account and project that should own the shared entries.

Authenticate A Developer Machine#

bash

once auth login --provider tuist

Once reuses the stored session for cache reads and writes. Remove it when the machine should no longer access the provider:

bash

once auth logout --provider tuist

Verify A Shared Result#

Use two machines with the same once.toml, script, and input. On the first machine, run the example scripted workflow:

bash

./scripts/greet.sh

The first run reports a cache miss. On a second machine whose local cache has not seen this result, create the same message.txt and run the script again:

bash

printf 'hello from Once\n' > message.txt
./scripts/greet.sh

The second machine should report a cache hit and restore build/greeting.txt. Because its local cache has no matching entry, the hit confirms that Tuist supplied the recorded result.

Authenticate Automation#

Choose the authentication method supported by the environment.

Account Token#

Set TUIST_TOKEN to a Tuist account token with cache access when the runner can store a long-lived secret:

bash

TUIST_TOKEN=tuist_... ./scripts/greet.sh

OpenID Connect#

Use OpenID Connect on a supported automation runner. Log in without opening a browser before running cacheable commands:

bash

once auth login --provider tuist --no-browser
./scripts/greet.sh

On GitHub Actions, grant identity-token permission before the login step:

yaml

permissions:
  id-token: write
  contents: read

steps:
  - uses: actions/checkout@v6
  - run: once auth login --provider tuist --no-browser
  - run: ./scripts/greet.sh

The resulting session remains available for the rest of the job.

Next#

Add an execution provider when the command should also run away from the current computer. Read Memory to inspect the status, cache state, and action identity that Once records for each run.