Skip to main content
Version: latest

Set up the example

The examples on the tool pages run against one short document, acme-notes.md. Load it into a graph once, and every example on the next pages works with the same objects: store, embedder, scope, and the helper short.

The names in the document are invented.

The document​

Save this text as acme-notes.md.

# Acme Robotics engineering notes

## Route planning

Maria Chen works for Acme Robotics and wrote the Orion Planner. The Orion Planner schedules delivery routes for the warehouse robots in the Berlin warehouse. It replans a route within two seconds when an aisle is blocked.

## Dependencies

The Orion Planner depends on Orion Maps, a map service that Daniel Okafor maintains. Daniel Okafor also works for Acme Robotics. Orion Maps supplies the floor plans that the planner uses to find free aisles.

| Component | Owner | Purpose |
| --- | --- | --- |
| Orion Planner | Maria Chen | Plans delivery routes |
| Orion Maps | Daniel Okafor | Serves warehouse floor maps |
| Fleet Monitor | Priya Raman | Tracks robot battery and health |

## Data retention

Robot logs are kept for 30 days and then deleted. A customer can ask Acme Robotics to erase their delivery data. After the request is verified, the data is deleted within 10 business days.

Load the document​

You need a running Neo4j database, the NEO4J_* variables in a .env file, and the neo4j and extract extras. Ingest documents describes each step of Graph.add.

import asyncio
from pathlib import Path

from agrag.common.data_models import DEFAULT_SCHEMA
from agrag.embedding import build_embedder
from agrag.graphdb import build_graph_store
from agrag.ingestion import Graph
from agrag.ingestion.extract import GlinerExtractor


def short(evidence_id: str) -> str:
    if len(evidence_id) <= 24:
        return evidence_id
    return evidence_id[:14] + "..." + evidence_id[-8:]


async def main() -> None:
    store = build_graph_store("neo4j")
    embedder = build_embedder("ibm-granite/granite-embedding-small-english-r2")
    graph = await Graph.open(
        schema=DEFAULT_SCHEMA,
        graph_store=store,
        embedder=embedder,
        extractor=GlinerExtractor(),
    )
    try:
        result = await graph.add(source="acme-notes.md")
        print(result.ingestion)
        scope = frozenset({str(Path("acme-notes.md").resolve())})
        ...  # the example snippets go here
    finally:
        await store.close()


asyncio.run(main())

The output of the load:

documents=1 sources=1 skipped=0 unchanged=0 quarantined=0 quarantined_items=[]

scope is the set of document keys that the tools may read. The key of a file is its absolute path, so the set holds the one document that you loaded. short shortens an evidence id for printing, because a full id is long.

What the graph holds​

The loader cut the document into three chunks. A chunk is the unit that search returns.

ChunkHolds
...3e66aa77The title, the "Route planning" section, and the "Dependencies" section without its table
...12cd16fbThe table in "Dependencies"
...d7c4d6efThe "Data retention" section

The extractor found these entities:

LabelNames
PersonMaria Chen, Daniel Okafor, Priya Raman
OrganizationAcme Robotics, Orion
SoftwareOrion Planner, Orion Maps, Fleet Monitor
Datasetfloor plans
Productwarehouse robots, robot battery

It also found the relations Maria Chen AUTHORED Orion Planner and Maria Chen WORKS_FOR Acme Robotics.

The extractor misses and invents some facts

A small local model did the extraction. It created an entity named Orion from a word that appears only inside the names of the planner and the map service, and it did not record that the planner depends on Orion Maps. The examples show these results as they are.

Your ids, paths, and scores differ from the ones on the next pages. An id depends on the path of the file and on its content, so the same file at the same path gives the same ids.