Node System

Nodes are the building blocks of every Salpa workflow. Each node represents a single computational step — from reading a PDB file to running a full GROMACS simulation.

How nodes work

A node has:

Inputs — Data it receives from upstream nodes (files, parameters, sequences)
Outputs — Data it produces for downstream nodes
Configuration — User-settable parameters (file paths, options, thresholds)
Execution logic — The actual computation (Python code)

When you connect nodes on the canvas, you define a data flow pipeline. Outputs from one node become inputs to the next.

Node package structure

Every node follows a two-level architecture:


my_node/
├── meta.toml      # Metadata: name, version, category, hashtags
├── node.py        # Salpa integration (inputs, outputs, UI options)
├── core.py        # Pure science code (no Salpa dependencies)
├── pixi.toml      # Python environment (optional — triggers isolation)
├── README.md      # Documentation
├── demo_data/     # Example input files
└── tests/         # Unit tests

Why two files?

core.py contains your actual computation — pure Python, no framework dependencies. You can test and run it independently.
node.py wraps core.py for Salpa — defines inputs, outputs, configuration UI, and execution flow.

This separation means your science code stays clean and reusable outside of Salpa.

Execution strategies

Salpa automatically selects the right execution strategy for each node:

Strategy	When	What happens
In-process	No `pixi.toml` present	Runs directly in the worker process. Fast, no isolation.
Pixi subprocess	`pixi.toml` present	Creates an isolated Python environment and runs in a subprocess. Full dependency isolation.
Container	Coming soon	Docker/Podman execution for maximum portability and reproducibility.
HPC / SLURM	Contact for support	Remote cluster execution via SSH. Vendor-dependent configuration.

Auto-detection: If your node directory contains a pixi.toml file, Salpa automatically uses the isolated subprocess strategy. Otherwise, it runs in-process.

Available nodes

Local nodes

Installed from the marketplace or built-in:

Node	Category	Description
Hello World	Utility	Simple test node for verifying execution
PDB-FASTA BioPython	Analysis	Extract FASTA sequences from PDB files
PDB2PQR	Preparation	Add hydrogens and assign charges to PDB structures
GROMACS MD Run	Simulation	Run molecular dynamics with GROMACS
ESMFold Prediction	Structure	Local protein structure prediction with ESMFold
AutoDock Vina	Docking	Molecular docking simulations

Cloud nodes

Run on remote GPU hardware (requires sign-in):

Node	GPU	Description
ESMFold (Cloud)	A100	Protein structure prediction via ESMFold
Boltz-2 (Cloud)	H100	Multi-chain structure prediction via Boltz-2

Naming conventions

Salpa uses consistent naming across different contexts:

Context	Convention	Example
Display name	Title Case with spaces	”GROMACS MD Run (Local)“
Package name (meta.toml)	kebab-case	`gmx-mdrun-local`
Directory name	snake_case	`gmx_mdrun_local/`
Python class	PascalCase	`GmxMdrunLocal`

Next steps

Build Custom Nodes — Create your own computational nodes
Cloud Computing — Run nodes on cloud GPUs
Getting Started — Install and run your first workflow