Skip to content

Superstep documentation

Superstep is a self-hosted platform for building LangGraph workflows in the browser and running them as services. A workflow is a JSON document: the FastAPI backend validates it, compiles it into a LangGraph StateGraph, and runs it with per-node events. The same workflow can be started over HTTP, on a cron schedule, or paused for a human decision.

Quickstart

git clone https://github.com/quinguyen1502/superstep.git
cd superstep && ./dev.sh

Open http://localhost:5173 and create an account with Sign up. LLM nodes answer from an offline mock until you add a model credential, so nothing else is needed to follow the quickstart: build a workflow, run it, approve it, and call it over HTTP in about ten minutes.

Explore the docs

Components

Part Built on Role
Editor React, React Flow, shadcn/ui Draws the graph and writes the workflow JSON; streams run events over WebSocket
API FastAPI, SQLAlchemy, Alembic Stores workflows and versions, validates and compiles them, serves REST and WebSocket
Engine LangGraph StateGraph Runs the graph in supersteps and checkpoints the state after each one
Worker In-memory queue or Redis Runs API-triggered, scheduled and recovered runs, and webhook deliveries
Storage SQLite, or Postgres with pgvector Workflows, runs, checkpoints, credentials and knowledge-base vectors
Sandbox Docker Python in Tool nodes and imported LangGraph repos
Telemetry OpenTelemetry One trace per run and one span per node, to OTLP or LangSmith

See Architecture for the request paths and the guarantees around retries and recovery.