Blacknode

The visual workflow builder where AI agents build the workflow.

Blacknode turns agent intent into typed, visible, runnable workflows. Agents get a structured control surface through MCP, HTTP, and WebSocket APIs instead of guessing JSON, and users get a live graph they can inspect, run, replay, and export.

Workflows export to plain Python, class-based Python, LangGraph, CrewAI, AutoGen, OpenAI Swarm, and an NVIDIA Agent Stack manifest, with NVIDIA NIM and AI-Q/NeMo Agent Toolkit workflow paths built in.

Demo1.mp4

Demo2.mp4

Start Here

New users should begin with the Beginner Walkthrough.

It shows the exact commands to run, buttons to press, templates to open, results to expect, NVIDIA NIM paths, MCP setup, framework export, Docker Compose, custom nodes, run history, and troubleshooting.

Why Blacknode

Chat agents are good at intent and iteration. They are weak at showing durable workflow state. Blacknode gives agents a typed workflow editor: they can create nodes, connect ports, validate the graph, run it, debug failures, replay the execution, and export the result as code.

Feature Map

Feature	What it gives you	Read more
Visual workflow editor	Build and inspect typed node graphs with visible execution state.	Beginner Walkthrough
Agent control surface	MCP, HTTP, and WebSocket APIs for agents to create, connect, validate, run, save, inspect, and export workflows.	Agent Guide, MCP Quickstart
NVIDIA workflow surface	Hosted NIM, local NIM launch planning, NIM benchmarks, AI-Q/NeMo Agent Toolkit integration, and streamable HTTP MCP.	NVIDIA Mission Control, NVIDIA NIM Demo, Blacknode and NVIDIA AI-Q
Typed ports and validation	Text, Int, Float, Bool, List, Dict, Embedding, Fn, Model, Number, Any, cycle checks, and MCP repair suggestions.	Workflow Schema, Agent Skill
Run history and replay	Event logs, model calls, tool calls, node timings, final values, and errors.	Beginner Walkthrough, Presentation Checklist
Custom nodes	Persistent editor-created nodes, Python decorator nodes, auto-discovery, and community node packs.	Custom Nodes
Python round-trip	Export readable Python, import it back into the editor, and live-sync Python runs into replay.	Python Round-Trip
Framework export	Turn a visual graph into Python, LangGraph, CrewAI, AutoGen, OpenAI Swarm, or an NVIDIA Agent Stack manifest.	Framework Export
Self-hosted deployment	Run the editor, backend, and HTTP MCP server locally, on a VM, or in an on-prem demo stack.	Docker Compose, Docker Publishing

NVIDIA Agent Stack

Blacknode complements NVIDIA AI-Q and NeMo Agent Toolkit by giving agent harnesses a visual workflow surface. AI-Q can research and reason over enterprise data; Blacknode turns agent intent into typed, visible, runnable workflows through MCP.

Blacknode is the visual workflow editor for the NVIDIA agent stack.

See Blacknode and NVIDIA AI-Q and NVIDIA Mission Control.

Documentation

First Run

Guide	Use it for
Beginner Walkthrough	Step-by-step setup, editor use, CLI checks, NVIDIA workflows, MCP, Docker, and troubleshooting.
Presentation Checklist	Fast demo order with actions, expected proof, and feature checkpoints.
MCP Quickstart	Connecting Blacknode to an MCP client.
MCP Test Prompts	Copy-paste prompts for proving agent workflow control.

NVIDIA

Guide	Use it for
NVIDIA NIM Demo	Hosted NVIDIA NIM demo path through MCP and the editor.
NVIDIA Mission Control	NVIDIA nodes, templates, local readiness, local NIM launch, and benchmark flow.
Blacknode and NVIDIA AI-Q	Positioning Blacknode beside AI-Q and using streamable HTTP MCP.

Deployment

Guide	Use it for
Docker Compose	Running the editor, backend, and HTTP MCP server as a self-hosted stack.
Docker Publishing	Publishing prebuilt server/editor images to GHCR and running without local builds.

Workflow Reference

Guide	Use it for
Workflow Schema	The saved workflow JSON format.
Workflow JSON Schema	Machine-readable schema for validation and tooling.
Framework Export	Exporting workflows to Python, LangGraph, CrewAI, AutoGen, Swarm, REST, and WebSocket control.
Python Round-Trip	Export Python, import Python back into the editor, and live-sync Python runs into replay.
Custom Nodes	Persistent editor-created nodes, auto-discovery, community node packs, and node library extension.
Agent Guide	How agents should create and modify Blacknode workflows.
Blacknode Skill	Agent skill instructions for workflow creation, validation, running, and export.

Demos

Demo	What it shows
MCP + NVIDIA NIM preview	Claude opens, organizes, and cooks an NVIDIA NIM workflow through MCP.
Run workflow live replay	The editor runs a visible graph with live node highlights and run replay.

Visuals

Preview	Link
MCP + NVIDIA NIM editor demo	docs/images/blacknode-mcp-nim-editor-demo.png
Claude Desktop MCP connector	docs/images/blacknode-mcp-claude-connector.png
Research pipeline template	docs/images/blacknode-research-pipeline.png
Light theme	docs/images/blacknode-light-theme.png
Dark theme	docs/images/blacknode-dark-theme.png

Project Map

Path	Purpose
python/blacknode/	Python workflow runtime, node registry, providers, CLI, and MCP server.
editor-server/	FastAPI backend for the visual editor, cook API, workflows, and runs.
editor/	React visual workflow editor.
templates/	Tracked starter workflows.
workflows/	Local saved workflows, ignored by git.
docs/	Walkthroughs, integration guides, workflow schema, and demo assets.
docker-compose.yml	Self-hosted editor, backend, and streamable HTTP MCP stack.
crates/	Rust crates and no-server CLI.

License

Blacknode is licensed under the Apache License 2.0. See LICENSE for the full license text.