bpmn2neo 0.2.4

Load BPMN 2.0 diagrams into Neo4j and create node embeddings for Graph-RAG.

bpmn2neo

Load BPMN 2.0 diagrams into Neo4j and create node embeddings for Graph-RAG.

---

1) Overview

bpmn2neo ingests BPMN 2.0 (.bpmn) files, persists a clean BPMN knowledge graph in Neo4j 5.x, then generates hierarchical embeddings for Graph-RAG.
The library is production-oriented: clean configuration via Pydantic .env, and a minimal public API.

• Parser & Loader: turn .bpmn XML into a normalized Neo4j schema.
• Embedding Orchestrator: builds node-level context and embeddings in a top-down order (Collaboration → Participant → Process → Lane → FlowNode).
• Graph-RAG ready: each node stores explanatory text and vector.

---

2) Key Features

BPMN → Neo4j Loader

• Robust XML parsing with clear phases (collaboration/participants → processes → lane → flownode → message flow/data objec/etc.).
• Builds nodes and relationships, mapping each BPMN element’s tag attributes to node properties.
• Structured logs for each phase, with try/except and error reporting.

👉 Schema details: see the guide Neo4j Graph Schema Guide

Hierarchical Context & Embedding

• For each node, the pipeline runs:
  Reader → Builder → ContextWriter → Embedder
• Generates text per node, then embeds using the configured embedding model.
• Order: FlowNode → Lane → Process → Participant → Model (Collaboration).
  This preserves hierarchy so parents can summarize/aggregate children.

👉 Embedding details: see the guide BPMN Embedding Rules Guide

---

3) Usage

Installation

pip install bpmn2neo

setting .env

# =========================
# bpmn2neo .env
# =========================
# --- Neo4j (REQUIRED) ---
B2N_NEO4J__URI="YOUR_URI"        # or neo4j+s://<host>:7687
B2N_NEO4J__USERNAME="YOUR_USERNAME"
B2N_NEO4J__PASSWORD="YOUR_PASSWORD"      # if you prefer keyring, comment this and use PASSWORD_ALIAS below
B2N_NEO4J__DATABASE="YOUR_DATABASE"

# Optional Neo4j tweaks:
# B2N_NEO4J__PASSWORD_ALIAS=bpmn2neo/neo4j      # resolve secret via OS keyring
# B2N_NEO4J__LOG_LEVEL=INFO                     # INFO/DEBUG/WARN/ERROR

# --- OpenAI (REQUIRED for embeddings/text) ---
B2N_OPENAI__API_KEY="YOUR_API_KEY"     # if you prefer keyring, comment this and use API_KEY_ALIAS below
B2N_OPENAI__EMBEDDING_MODEL="text-embedding-3-large"

# Optional OpenAI tweaks:
# B2N_OPENAI__API_KEY_ALIAS=bpmn2neo/openai     # resolve secret via OS keyring
# B2N_OPENAI__EMBEDDING_DIMENSION=3072
# B2N_OPENAI__TRANSLATION_MODEL=gpt-4o-mini
# B2N_OPENAI__TEMPERATURE=0.2
# B2N_OPENAI__MAX_TOKENS_FULL=600
# B2N_OPENAI__MAX_TOKENS_SUMMARY=200
# B2N_OPENAI__MAX_RETRIES=3
# B2N_OPENAI__TIMEOUT=60
# B2N_OPENAI__LOG_LEVEL=INFO

# --- Container metadata  ---
B2N_CONTAINER__CREATE_CONTAINER=True # default : true
B2N_CONTAINER__CONTAINER_TYPE="YOUR_CONTAINER_TYPE"
B2N_CONTAINER__CONTAINER_ID="YOUR_CONTAINER_ID"
B2N_CONTAINER__CONTAINER_NAME="YOUR_CONTAINER_NAME"

# --- Runtime (optional) ---
# B2N_RUNTIME__LOG_LEVEL=INFO                  # global log level
# B2N_RUNTIME__PARALLELISM=1                   # worker parallelism
# B2N_RUNTIME__BATCH_SIZE=64                   # embedding batch size
# B2N_RUNTIME__CACHE_DIR=/tmp/bpmn2neo_cache   # local cache dir if you need one
# B2N_RUNTIME__DRY_RUN=false                   # true: do not write to DB
# B2N_RUNTIME__FAIL_FAST=false                 # true: stop on first error

Python API (recommended)

from bpmn2neo import load_bpmn_to_neo4j, create_node_embeddings, load_and_embed
from bpmn2neo.settings import Settings, Neo4jSettings, OpenAISettings

# Option A) reads from .env
s = Settings()  # reads .env via pydantic-settings

# Option B) by each Settings object
s2 = Settings(
    neo4j=Neo4jSettings(
        uri="bolt://localhost:7687",
        user="neo4j",
        password="your_password",
        database="neo4j",
    ),
    openai=OpenAISettings(
        api_key="sk-...",  # or api_key_alias="openai/default" if using keyring
    ),
)

# 1) Load only
model_keys = load_bpmn_to_neo4j(
    bpmn_path="./data/bpmn/Order Process for Pizza.bpmn",
    settings=s,
)
print("model_keys:", model_keys)

# 2) Embedding only
# mode="all"   : full hierarchy (FlowNodes → Lanes → Process → Participant → Model)
# mode="light" : FlowNodes only (faster iteration)
create_node_embeddings(model_key=model_keys[0], settings=s, mode="all")
create_node_embeddings(model_key=model_keys[0], settings=s, mode="light")

# 3) Pipeline (load + embed)
result = load_and_embed(
    bpmn_path="./data/bpmn/Order Process for Pizza.bpmn",
    settings=s,
    mode="light",  # or "all"
)
print("final model_key:", result["model_key"])

Requirements

• Python 3.10+
• Neo4j 5.x (Bolt reachable).
  Ensure the user has MATCH/CREATE/MERGE/SET/DELETE and index/constraint privileges.
• OpenAI API key (for embeddings & text generation).

---

4) Project Structure

bpmn2neo/
├─ NEO4J_SCHEMA.md         # Neo4j graph schema documentation
├─ EMBEDDING_RULES.md      # Embedding rules and pipeline documentation
├─ src/bpmn2neo/
│  ├─ config/
│  │  ├─ exceptions.py        # Domain exceptions (Config/Neo4j/etc.)
│  │  ├─ logger.py            # Structured logger (JSON-friendly)
│  │  └─ neo4j_repo.py        # Thin Neo4j driver wrapper + helpers
│  ├─ embedder/
│  │  ├─ builder.py           # Build signals/texts from Reader context
│  │  ├─ context_writer.py    # Ask LLM to craft node explanations
│  │  ├─ embedder.py          # Vectorize and persist embeddings
│  │  ├─ orchestrator.py      # Orchestrates Reader→Builder→Writer→Embedder
│  │  └─ reader.py            # Read graph context per node for embedding
│  ├─ loader/
│  │  ├─ loader.py            # High-level load flow (schema ensure + write)
│  │  └─ parser.py            # BPMN XML → nodes/relationships
│  ├─ settings.py             # Pydantic-based config (.env)
│  ├─ cli.py                  # Optional CLI entry
│  └─ __init__.py             # Public API: load, embed, pipeline
└─ pyproject.toml

---

5) License

Licensed under Apache License 2.0.
See LICENSE.
Notes:

• No trademark rights are granted.
• Contributions (if any) are accepted under the same license.
• Patent grant/termination follows Apache-2.0 terms.

---

Configuration & Tips

• Settings resolution: Settings() reads environment variables (and .env) using pydantic-settings. Keys are prefixed (e.g., B2N_NEO4J__URI).
• Model key: if not provided, the loader derives it from the BPMN filename (stem).
• Security: you may use keyring aliases instead of plain secrets (see comments in settings.py).

---

Troubleshooting

• Cannot connect to Neo4j

  • Verify B2N_NEO4J__URI (neo4j://host:7687 or neo4j+s://... for Aura).
  • Check firewall, DB auth, and database name.

• OpenAI errors

  • Ensure B2N_OPENAI__API_KEY set; consider rate limits and retries.

---