Use cases
Tubeless runs TypeScript functions as steps in a pipeline. Here are some ways to use it, with example workflows and links to working code.
01 / CI/CD
Build and release software
A release pipeline can build an artifact, run checks, and publish once those checks pass. The same code runs locally and in CI. Tubeless uses this setup for its own CI, where it runs linting, type checks, tests, documentation checks, and package verification.
Release previews
Preview a target and its prerequisites without executing steps. Mark publication steps to skip during dry runs.
Debugging failed builds
Record step results, timings, and logs. Export an NDJSON trace as a CI artifact to inspect after the job ends.
Build, test, and deploy
Illustrative workflow: A build fans out into tests and packaging. Both branches join at a release gate before deployment.
02 / ML flows
Prepare data and evaluate models
Use a pipeline to prepare a dataset, submit a training job, and evaluate the resulting model. Training can run on a remote service while preparation and evaluation use your existing functions.
Comparing models
Run the same evaluation pipeline for several datasets or model candidates. Limit how many run at once and track progress for each.
Checking training results
Validate the training service’s response with a schema. An evaluation step can reject models that fail your criteria and block registration.
Compare three model experiments
Illustrative workflow: A dataset fans out to three model experiments. Their results converge for evaluation, and the chosen model is registered.
03 / LLM workflows
Process documents with an LLM
A document processing job might retrieve context, call a model, and check the response before saving it. Write the prompts and provider calls as ordinary functions; Tubeless tracks their progress and results.
API limits and retries
Retry and rate-limit helpers wrap provider calls. When processing documents in parallel, you can cap concurrency and pass cancellation through to each request.
Checking model output
Check structured responses against a schema before saving them. A custom dry-run handler can return a sample response without making a paid model call.
Answer a question with retrieved context
Illustrative workflow: A question starts retrieval and a tool call. Retrieved context, tool results, and the original question all feed a model, which produces an answer.
04 / Data pipelines
Import and transform data
Split an import into steps for reading, normalizing, enriching, and writing records. Each step declares which outputs it needs, so TypeScript can check the data passed between them.
Batch progress and failures
Report how many items have completed and identify which child pipeline items failed. Checkpointed batch helpers can track completed work for a later run.
Dry runs
Configure write steps to skip during a dry run, or return a preview without changing data. Recorded runs include logs and timings you can inspect in the local Studio.
Combine data from several sources
Illustrative workflow: Database rows, API responses, and files merge into a normalization step. The cleaned data branches out to a warehouse and a report.
05 / Operational workflows
Run maintenance tasks
Register maintenance scripts as named commands with declared inputs. Teammates can see what’s available, preview a task, and follow its progress when they run it.
Shared commands
Group pipelines with defineProject to share them with your team. Schema-backed pipelines run from the terminal or launch through the local Studio.
Troubleshooting a run
Record runs to see which steps completed before a failure or cancellation. For commands that change data, add a dry-run handler so teammates can review the proposed changes first.
Back up, repair, and verify
Illustrative workflow: A maintenance request branches into a backup and preflight checks. Both must finish before a repair, which branches into verification and an audit record.
Getting started
The getting started guide walks through a small pipeline. If you’re adapting an existing script with a coding agent, install the Tubeless skill pack.