#wlow

wlow is a DAG-based workflow orchestrator. You define a graph of tasks, submit it, and wlow runs each task in the right sandbox — with dependency tracking, retries, and a final result.

Processors run in three execution tiers:

Status: pre-1.0, actively developed.

---

#How it works

You write a processor, push it to wlow, and run it anywhere — the processor just needs NATS connectivity and the wlow binary (for process/WASM) or the wlow runner image (for microVMs).

                     wlow.processor.sandbox.>
Orchestrator ──────────────────────────────────► wlow start (your container)
     ▲                                                  └─ spawns your script/binary per task
     │
     │  workflow.reply.<id>
     └──────────────────────────────────────────── result

For microVM runtimes the runner is our image (with Firecracker inside), not yours.

---

#Install

# Linux / macOS — one-liner
curl -fsSL https://raw.githubusercontent.com/wlow/wlow-core/main/install.sh | sh

# Homebrew
brew install wlow/tap/wlow

# Go install (requires Go 1.23+)
go install github.com/wlow/wlow-core/cmd/wlow@latest

# Direct download — https://github.com/wlow/wlow-core/releases

Verify: wlow version

---

#Quickstart

#1. Start NATS and the control plane

nats-server --js &
make wlow
./bin/wlow start --control-plane

#2. Scaffold a processor

./bin/wlow new my-proc
# creates my-proc/processor.py and a Dockerfile

#3. Start it — no push needed

# Any machine with wlow + python3 + NATS connectivity:
./bin/wlow start --id my-proc --cmd "python3 my-proc/processor.py"

--cmd auto-registers the processor manifest in NATS and starts consuming tasks immediately. No wlow push required for process or Go SDK processors.

#4. Submit a workflow

client, _ := sdk.NewClient(sdk.ClientConfig{NATSUrl: "nats://localhost:4222"})
wf, _ := workflow.NewBuilder("job-1").
    AddTask("step", workflow.Task{
        ProcessorID: "my-proc", ProcessorVersion: "latest",
        Input: map[string]any{"text": "hello world"},
    }).Build()
result, _ := client.SubmitAndWait(ctx, wf, time.Minute)

---

#The wlow CLI

wlow start --control-plane   Start the control plane (orchestrator)
wlow start --id P --cmd CMD  Start a process processor — no push needed
wlow start --runtimes wasm   Start a WASM processor runner
wlow new <name>              Scaffold a new processor project
wlow push                    Register a WASM or microVM processor artifact
wlow prepare-snapshot        Prepare snapshot artifacts (run from a KVM host)
wlow benchmark               Timing tests

#When do you need wlow push?

---

#Deploying a process processor as a container

The container owns its runtime (python3, dependencies, the script). wlow start --cmd registers the processor manifest on startup — no prior push step.

# my-proc/Dockerfile
FROM python:3.12-slim

COPY my-proc/processor.py /app/processor.py
# RUN pip install your-dependencies

RUN curl -fL -o /usr/local/bin/wlow \
      https://github.com/wlow/wlow/releases/latest/download/wlow-linux-amd64 \
    && chmod +x /usr/local/bin/wlow

ENV NATS_URL=nats://nats:4222
ENTRYPOINT ["wlow", "start", "--id", "my-proc", "--cmd", "python3 /app/processor.py"]

Build and run:

docker build -t my-proc:latest .
docker run -e NATS_URL=nats://your-nats:4222 my-proc:latest

Scale by running more instances. They all share the same NATS task queue.

---

#MicroVM processors

For microvm or snapshot, you push a Dockerfile:

wlow push --id my-proc --runtime microvm \
  --path my-proc/Dockerfile --entrypoint python3,/app/processor.py \
  --registry ghcr.io/your-org/wlow-artifacts

Then deploy the wlow runner image on a KVM-capable host and it handles execution. See docs/runner-setup.md for KVM setup on GCP, AWS, and Linux workstations.

---

#Build

make wlow               # the CLI — start, new, push, prepare-snapshot, benchmark
make linux-amd64-bins   # all binaries cross-compiled for linux/amd64

---

#Documentation

---

#Performance (single runner, KVM)

---

#Repositories

The runtime container image bundles binaries from both wlow-core and wlow-runner.

#Contributing

CONTRIBUTING.md · SECURITY.md · Apache-2.0