Skip to content

framework-export.md

Latest commit

 

History

History
110 lines (84 loc) · 4.53 KB

framework-export.md

File metadata and controls

110 lines (84 loc) · 4.53 KB
Raw

Framework Export

Blacknode keeps the visual workflow as the source of truth and exports the same typed graph into other agent frameworks when you want to run or extend it outside the editor.

Editor

  1. Start Blacknode with .\start.bat on Windows or ./start.sh on macOS/Linux.
  2. Open or build a workflow that ends in an Output or OutputImage node.
  3. Press Export in the top bar.
  4. Choose Plain Python, Python Class, LangGraph, CrewAI, AutoGen, OpenAI Swarm, or NVIDIA Agent Stack.
  5. The generated file downloads from the browser.

Editor export infers an entrypoint for common multi-output graphs. It preserves an explicit workflow entrypoint when one exists, otherwise it prefers visible display nodes such as overlay_out.image, reason_dashboard_out.image, and other OutputImage nodes before falling back to an Output.value node. This keeps live robotics/vision dashboards exportable even when the graph contains several output panels.

CLI

Export a LangGraph StateGraph:

blacknode export-framework templates\text-pipeline.json --target langgraph --output workflow.langgraph.py

Export plain Blacknode Python:

blacknode export-framework templates\text-pipeline.json --target python --output workflow.python.py
blacknode export-framework templates\text-pipeline.json --target python-class --output workflow.class.py

Plain Blacknode Python exports include a runtime bootstrap. A downloaded export can run from another folder, and BLACKNODE_HOME=/path/to/Blacknode overrides which local checkout supplies the runtime and .venv dependencies.

When a plain Python export contains live ROS 2, CV2, vision reasoning, or ROS 2 run nodes, it keeps running after printing the stream URL. Use Ctrl-C to stop the helper processes cleanly.

Import a Blacknode Python export back into workflow JSON:

blacknode import-python workflow.python.py --output imported.workflow.json
blacknode import-python workflow.langgraph.py --output imported-langgraph.workflow.json

The editor Import button and canvas file drop use the same importer, so Blacknode-generated Python and LangGraph files can be reopened as visual workflow tabs.

For CLI exports, workflow JSON should still include an explicit entrypoint when it has multiple Output nodes. The editor adds one automatically for the current graph before calling the same exporter.

Export framework maps:

blacknode export-framework templates\text-pipeline.json --target crewai --output workflow.crewai.py
blacknode export-framework templates\text-pipeline.json --target autogen --output workflow.autogen.py
blacknode export-framework templates\text-pipeline.json --target swarm --output workflow.swarm.py
blacknode export-framework templates\text-pipeline.json --target nvidia-agent-stack --output workflow.nvidia-agent-stack.py

HTTP

List targets:

Invoke-RestMethod http://127.0.0.1:7777/export/frameworks

Export the current editor graph:

Invoke-RestMethod `
  -Method Post `
  -Uri http://127.0.0.1:7777/export/framework `
  -ContentType application/json `
  -Body '{"target":"langgraph"}'

Direct non-MCP workflow API:

Endpoint Purpose
GET /api/workflows/current Current workflow JSON plus validation.
POST /api/workflows/current/nodes Add a node with type_name, params, and pos.
POST /api/workflows/current/edges Connect two ports.
GET /api/workflows/current/validate Validate the current workflow.
POST /api/workflows/current/run Cook a selected node and port.
POST /api/workflows/current/export Export the current workflow to a framework target.
POST /api/workflows/current/import-python Import a Blacknode Python export into workflow JSON.
WS /api/workflows/current/ws WebSocket state, validation, and export actions.

Target Behavior

Target Output
python Runnable Blacknode bn.Graph() script with round-trip metadata and optional live sync.
python-class Runnable class-based Blacknode script for cleaner embedding.
langgraph LangGraph StateGraph with START, END, node functions, edges, and final result print.
crewai CrewAI task descriptors mapped from Blacknode nodes and upstream context.
autogen AutoGen agent descriptors with handoff targets from graph edges.
swarm OpenAI Swarm-style handoff descriptors from graph edges.
nvidia-agent-stack AI-Q, NeMo Agent Toolkit, and NIM integration manifest with MCP commands and readable workflow steps.