Skip to content

Tolquane

Parallel programming with composable building blocks, in Python.

Nodes speak on channels. Pipelines, farms and all-to-all blocks compose them, and the same graph runs on threads, in child processes, across machines, or on one thread for debugging. Tolquane is the successor of BBFlow, a Java implementation of the FastFlow building blocks, rebuilt from scratch to be simple to use and impossible to hang.

import tolquane as tq

@tq.source
def numbers():
    yield from range(1, 101)

@tq.node
def double(x: int) -> int:
    return x * 2

@tq.sink
def show(x: int) -> None:
    print(x)

tq.run(numbers >> tq.farm(double, workers=4, ordered=True) >> show)

A function is a node. >> builds a pipeline. A farm runs copies of a node in parallel. Return tq.SKIP to drop an item; None is an ordinary value; a generator yields many.

What you get

  • Blocks: sources, nodes, sinks, farms with round-robin, on-demand, broadcast, scatter and keyed emitters and first-come, round-robin, gather and ordered collectors, node fusion, all-to-all, feedback loops that terminate by rule.
  • Runtimes: threads (full parallelism on free-threaded Python), processes for CPU-bound work on GIL builds, TCP between hosts from a deploy file, coroutine pools for network-bound stages, and a deterministic single-thread runtime for tests.
  • No hangs: every edge is bounded, every node waits on one inbox, errors cancel the run and name the node, and a watchdog reports deadlocks by name.
  • An AI builder: tolquane build "..." writes, checks and runs a flow from a sentence with your own Claude or GPT key.
  • A GUI: tolquane web opens a canvas of the blocks with the Python beside it, live runs, a history and schedules.

Try Tolquane Web without installing anything on the live demo (sign in as demo with the password tolquane). Install with pip install tolquane. Pure Python, 3.11 or newer, no required dependencies. Then read the tutorial, keep the API card open, and pick a runtime.

Tolquane Web

pip install "tolquane[web]" and tolquane web open a local page for the flows in one directory: the blocks on a canvas, the Python one click away and editable both ways, runs with per-node counts and tapped items as they happen, a run history, cron schedules and the AI builder in a side panel. The file is still a plain flow.py, so anything made there runs with python flow.py wherever Tolquane is installed. Since 1.3 it also has accounts with roles, parameters and environment variables for a run or a schedule, a git history of every flow, and webhooks, mail and retries when a schedule ends. The user guide is the tour, and Deploying Tolquane Web covers a shared machine, a systemd service, Docker and a reverse proxy.