callbag-recharge vs Apache Airflow

Both orchestrate data pipelines with DAG semantics. callbag-recharge is lightweight, reactive, runs anywhere, and requires no infrastructure.

At a Glance

Feature	Airflow	callbag-recharge
Language	Python	TypeScript
Infrastructure	Scheduler + Workers + DB + UI	None (runs in-process)
Runs in browser	No	Yes
DAG definition	Python decorators / YAML	pipeline() + step() — TypeScript
Scheduling	Built-in scheduler	fromCron() — zero-dep cron parser
Execution model	Polling (check DB)	Reactive (push-based)
Persistence	PostgreSQL/MySQL	checkpoint() — file, SQLite, IndexedDB
Human-in-the-loop	Manual approval sensor	gate() — native, reactive
Monitoring	Web UI + logs	Inspector + executionLog() + toSSE()
Retry	Task-level retry	retry(), withRetry() — composable
Latency	Seconds (polling + DB)	Microseconds (in-process, push)

The Key Difference

Airflow is a platform — you deploy it, manage it, and build on it. callbag-recharge is a library — npm i @callbag-recharge/callbag-recharge and compose pipelines in your existing TypeScript codebase.

// Airflow (Python)
@dag(schedule='0 9 * * *', start_date=datetime(2024, 1, 1))
def daily_pipeline():
    data = fetch_data()
    transformed = transform(data)
    save(transformed)

// callbag-recharge (TypeScript)
const daily = fromCron('0 9 * * *')
const data = pipe(daily, exhaustMap(() => fromPromise(fetchData())), retry(3))
const transformed = derived([data], () => transform(data.get()))
effect([transformed], () => save(transformed.get()))

What Airflow Lacks

1. Browser / Edge execution

Airflow requires a server. callbag-recharge runs in the browser, edge runtimes, serverless functions, or your laptop.

2. Reactive execution

Airflow polls a database to check if tasks are ready. callbag-recharge pushes values through the graph — microsecond latency vs seconds.

3. TypeScript native

Airflow is Python. If your app is TypeScript, you maintain two stacks. callbag-recharge is your app language.

4. Composability

Airflow tasks are isolated Python functions. callbag-recharge steps are reactive stores — compose them with derived(), pipe(), switchMap(), and the full operator set.

What Airflow Does Better

• Battle-tested at scale — production-proven for massive data pipelines (1000+ DAGs)
• Rich UI — web dashboard for DAG visualization, task logs, trigger management
• Ecosystem — 1000+ pre-built operators (AWS, GCP, Spark, Kubernetes, etc.)
• Multi-team coordination — RBAC, connection management, variable store
• Distributed execution — Celery/Kubernetes executors for horizontal scaling

When to Choose callbag-recharge

• Your pipeline is part of a TypeScript application (not a standalone data platform)
• You need browser-side or edge execution
• You want reactive (push-based) execution, not polling
• Infrastructure overhead of Airflow is too high for your use case
• You need human-in-the-loop (gate()) or real-time monitoring (toSSE())
• Your pipeline has < 100 steps and doesn't need distributed execution