Compose Python logic into data-flow pipelines — sync or async, run anywhere.
Write plain Python functions. Wire them with >>. Data flows visibly
between nodes — you can see what went in, what came out, what failed.
Unlike task schedulers where jobs are opaque units, lythonic tracks the
data itself.
uv add lythonicfrom lythonic.compose.namespace import Dag
def fetch(url: str) -> dict:
return {"source": url, "values": [1, 2, 3]}
def double(data: dict) -> dict:
return {**data, "values": [v * 2 for v in data["values"]]}
dag = Dag()
dag.node(fetch) >> dag.node(double)
# Run it — sync or async, doesn't matter
import asyncio
result = asyncio.run(dag(url="https://example.com"))
print(result.outputs) # {"double": {"source": "...", "values": [2, 4, 6]}}Data flow, not task flow. Each node receives typed data from upstream and passes results downstream. The DAG runner wires inputs to outputs by type — fan-out, fan-in, and map-reduce built in. Provenance tracking records what data flowed through each edge.
Compose freely. DAGs nest inside DAGs. dag.node(sub_dag) runs a
sub-DAG as a single step. dag.map(sub_dag) runs it on each element of
a collection, concurrently. Build small, reuse everywhere.
Run transparently. await dag() for a quick test. DagRunner with
provenance for production. lyth start for a long-running engine with
cron-triggered pipelines. Same code, different execution context.
Sync and async — mixed freely. Write sync functions, async functions, or both in the same DAG. Sync nodes run in a thread executor automatically.
- DAG composition —
>>wiring, callable DAGs, MapNode, CallNode @dag_factory— define reusable DAG templates as decorated functions- Triggers — cron-scheduled or push-triggered execution via
TriggerManager - Provenance — SQLite-backed tracking of runs, node executions, edge traversals
- Caching — per-callable SQLite cache with probabilistic TTL refresh
lythCLI —start,stop,run,fire,statuscommands- State — Pydantic-based SQLite ORM with schema management and multi-tenant support
- Portable data types —
FrameData,ExposureMatrix,SymmetricMatrix,KeyedVector: JSON in, numpy/pandas/polars/pyarrow out, no library required
Full documentation at walnutgeek.github.io/lythonic.