Skip to content

Workflow Builder

Visual pipeline builder for the JARVIS computational materials science ecosystem. Drag-and-drop 32 blocks across 9 categories (input, transform, DFT, force field/MLFF, GNN/ML, tight binding, quantum, analysis, output) to build computational pipelines. Choose from 8 pre-built templates or create custom workflows. Auto-generates ready-to-run Python scripts using jarvis-tools, ALIGNN, AtomGPT, SlakoNet, and Qiskit.

Open App

Overview

The Workflow Builder is a code generator, not a compute engine — it produces Python scripts that you run locally or on a cluster. The three-panel UI has a block palette (left), pipeline canvas (center), and live script preview (right). Each block maps to a Python code template that is composed into a complete script with proper variable wiring between blocks.

How It Works

No computation happens server-side. The app generates Python scripts from your pipeline definition. You run the generated .py file in your own environment with the required packages installed.

Block Categories

Category Blocks Description
Input jarvis_id, poscar_input, formula_input Load structures from JARVIS, POSCAR text, or formula
Transform supercell, vacancy, surface, heterostructure Structure manipulation
DFT vasp_relax, vasp_bands, vasp_optics, vasp_elastic, vasp_phonon, vasp_factory, qe_relax VASP and QE job generation
Force Field alignn_ff_opt, alignn_ff_md, alignn_ff_phonon, lammps_ff, chipsff_bench ML and classical force fields
GNN / ML alignn_predict, atomgpt_predict, atomgpt_generate, cfid_descriptors Property prediction and inverse design
Tight Binding slakonet_bands, wannier_tb, chipstb_bench TB band structure methods
Quantum vqe_electron VQE electronic structure via Qiskit
Analysis xrd_analysis, rdf_analysis, spacegroup_analysis Structural characterization
Output save_poscar, leaderboard_submit Export and benchmarking

Pre-built Templates

Template Pipeline
DFT Full JARVIS ID → VASP JobFactory (relax + bands + optics + elastic)
MLFF JARVIS ID → ALIGNN-FF optimize → MD → phonons
GNN Screen JARVIS ID → ALIGNN predict (formation energy, band gap, bulk modulus)
Surface Study JARVIS ID → surface slab → spacegroup analysis → XRD
Defect Study JARVIS ID → vacancies → ALIGNN-FF optimize
TB Bands JARVIS ID → SlakoNet bands → Wannier TB bands
Quantum VQE VQE electronic at X-point for Al
Inverse Design AtomGPT generate → ALIGNN validate → save POSCAR

Endpoints

GET /workflow/blocks — List all available blocks

Returns the full block definition catalog with parameters, categories, and I/O types.

curl "https://atomgpt.org/workflow/blocks" \
  -H "Authorization: Bearer sk-XYZ" \
  -H "accept: application/json"

GET /workflow/templates — List pre-built templates

Returns 8 template definitions with block sequences and default parameters.

curl "https://atomgpt.org/workflow/templates" \
  -H "Authorization: Bearer sk-XYZ" \
  -H "accept: application/json"

POST /workflow/generate — Generate Python script

Submit a pipeline definition (list of blocks with parameters) and receive a ready-to-run Python script.

curl -X POST "https://atomgpt.org/workflow/generate" \
  -H "Authorization: Bearer sk-XYZ" \
  -H "Content-Type: application/json" \
  -H "accept: application/json" \
  -d '{
    "name": "si_screening",
    "blocks": [
      {
        "type": "jarvis_id",
        "params": {"jid": "JVASP-1002", "dataset": "dft_3d"}
      },
      {
        "type": "alignn_predict",
        "params": {"model_name": "formation_energy_peratom"}
      },
      {
        "type": "alignn_predict",
        "params": {"model_name": "optb88vdw_bandgap"}
      },
      {
        "type": "alignn_ff_opt",
        "params": {"model": "alignnff_wt10", "fmax": "0.05", "steps": "200"}
      }
    ]
  }'

MLFF workflow example:

curl -X POST "https://atomgpt.org/workflow/generate" \
  -H "Authorization: Bearer sk-XYZ" \
  -H "Content-Type: application/json" \
  -H "accept: application/json" \
  -d '{
    "name": "mlff_workflow",
    "blocks": [
      {"type": "jarvis_id", "params": {"jid": "JVASP-1002", "dataset": "dft_3d"}},
      {"type": "alignn_ff_opt", "params": {"model": "alignnff_wt10", "fmax": "0.05", "steps": "200"}},
      {"type": "alignn_ff_md", "params": {"model": "alignnff_wt10", "temp": "300", "timestep": "1", "nsteps": "1000"}},
      {"type": "alignn_ff_phonon", "params": {"model": "alignnff_wt10", "dim": "2,2,2"}}
    ]
  }'
Field Type Description
name string Workflow name (used as script filename)
blocks list Ordered list of pipeline blocks
blocks[].type string Block type key (e.g. "jarvis_id", "alignn_predict")
blocks[].params dict Block-specific parameters (see block definitions)
blocks[].id string Optional custom block ID

Response:

Field Description
script Generated Python script as string
name Workflow name
n_blocks Number of blocks in pipeline
block_types Ordered list of block type names

Python Examples

import requests

response = requests.post(
    "https://atomgpt.org/workflow/generate",
    headers={
        "Authorization": "Bearer sk-XYZ",
        "accept": "application/json",
        "Content-Type": "application/json",
    },
    json={
        "name": "gnn_screen_si",
        "blocks": [
            {"type": "jarvis_id", "params": {"jid": "JVASP-1002", "dataset": "dft_3d"}},
            {"type": "alignn_predict", "params": {"model_name": "formation_energy_peratom"}},
            {"type": "alignn_predict", "params": {"model_name": "optb88vdw_bandgap"}},
            {"type": "alignn_predict", "params": {"model_name": "bulk_modulus_kv"}},
        ],
    },
)
data = response.json()
print(f"Generated {data['n_blocks']}-block script")
print(f"Pipeline: {' → '.join(data['block_types'])}")

with open(f"{data['name']}.py", "w") as f:
    f.write(data["script"])
print(f"Saved: {data['name']}.py")

import requests

response = requests.post(
    "https://atomgpt.org/workflow/generate",
    headers={
        "Authorization": "Bearer sk-XYZ",
        "accept": "application/json",
        "Content-Type": "application/json",
    },
    json={
        "name": "dft_full_pipeline",
        "blocks": [
            {"type": "jarvis_id", "params": {"jid": "JVASP-1002"}},
            {"type": "vasp_factory", "params": {
                "steps": "ENCUT,KPLEN,RELAX,BANDSTRUCT,OPTICS,ELASTIC",
                "functional": "optb88vdw",
            }},
        ],
    },
)
data = response.json()
with open("dft_full_pipeline.py", "w") as f:
    f.write(data["script"])

import requests

response = requests.get(
    "https://atomgpt.org/workflow/blocks",
    headers={"Authorization": "Bearer sk-XYZ"},
)
blocks = response.json()["blocks"]
for cat in ["input", "transform", "dft", "ff", "ml", "tb", "quantum", "analysis", "output"]:
    items = [k for k, v in blocks.items() if v["category"] == cat]
    if items:
        print(f"\n{cat.upper()}: {len(items)} blocks")
        for k in items:
            print(f"  {k:25s} {blocks[k]['label']}")

AGAPI Agent [WIP]

from agapi.agents import AGAPIAgent
import os

agent = AGAPIAgent(api_key=os.environ.get("AGAPI_KEY"))

# Generate MLFF workflow
response = agent.query_sync("Generate an MLFF workflow script for Si using ALIGNN-FF")
print(response)

References