Skip to main content

Langflow on Stepflow with a Single Command

· 4 min read
Ben Chambers
Ben Chambers
Co-creator of Stepflow

In our first Langflow on Stepflow post, we showed how to convert and execute Langflow flows on Stepflow. That demo required building the Stepflow binary from source, generating configuration files, and wiring everything together manually. It worked, but it wasn't exactly a one-liner.

Today, the stepflow-langflow CLI makes this dramatically simpler, thanks partly to the new stepflow-orchestrator package which includes the necessary binary to run locally. A single run command handles conversion, orchestrator startup, and execution — all in one step.

Installation​

pip install stepflow-langflow-integration

This pulls in the stepflow-orchestrator package automatically, which bundles the Stepflow server binary for your platform.

The run command​

The new run command supports two modes:

  • --local: Starts a local Stepflow orchestrator automatically using the stepflow-orchestrator package. No binary to build, no config to write.
  • --url: Connects to an already-running Stepflow server (local or remote).

Running locally​

To demonstrate, let's use the same Blog Content generation flow.json from our original POC post. Set your OpenAI API key and run:

export OPENAI_API_KEY="sk-..."

stepflow-langflow run flow.json '{}' --local

That's it — no tweaks, no config files, no binary to build. Langflow components that use load_from_db (like the Language Model's API key field) automatically resolve from environment variables at execution time, just as they do in Langflow itself.

The CLI:

  1. Converts the Langflow JSON to a Stepflow workflow
  2. Starts a local orchestrator with a Langflow component server
  3. Submits the workflow and waits for results
  4. Prints the output and shuts down

Compare this to the original POC, which required cloning the repo, building Rust from source, passing API keys via --tweaks, and running from a specific directory.

Running against a deployed server​

If you have a Stepflow server running (perhaps deployed to Kubernetes), point --url at it instead:

stepflow-langflow run flow.json '{"message": "Hello"}' --url http://stepflow.example.com:7840

The conversion happens client-side; only the final Stepflow workflow is submitted to the server. This means you can iterate on your Langflow flow locally and submit it to a production cluster when ready.

Batch execution​

To run the same flow across many inputs, put one JSON object per line in a JSONL file and use --batch:

stepflow-langflow run flow.json --batch inputs.jsonl --local

Where inputs.jsonl looks like:

{"message": "Write a haiku about coding"}
{"message": "Write a haiku about coffee"}
{"message": "Write a haiku about open source"}

The flow is converted and uploaded once, then all inputs are executed in a single run. Stepflow handles scheduling and parallelization automatically.

Dry run​

To inspect the converted Stepflow workflow without executing it:

stepflow-langflow run flow.json '{}' --local --dry-run

This prints the generated YAML so you can see exactly how Langflow components map to Stepflow steps.

How it works​

Under the hood, --local uses the stepflow-orchestrator Python package, which bundles the Stepflow server binary. The CLI:

  1. Converts the Langflow JSON to a Stepflow workflow using the built-in translator
  2. Starts a local StepflowOrchestrator with a config that includes the Langflow component server as a sub-process plugin
  3. Stores the workflow via the HTTP API (POST /api/v1/flows)
  4. Submits a run and waits for completion (POST /api/v1/runs with wait=true)
  5. Displays the results
  6. Stops the local orchestrator

For --url mode, steps 2 and 6 are skipped — the CLI talks directly to the remote server.

What's next​

We're continuing to improve the experience:

  • Streaming output: Follow step execution in real time
  • Flow caching: Skip re-uploading unchanged flows

If you're using Langflow and want to run your flows with distributed execution, parallel step processing, or Kubernetes-scale deployment, give stepflow-langflow run --local a try.