Adapter Contractv1.0

Graph / RAG Adapter Spec

Adapter contract for graph-based memory and RAG systems. These systems use knowledge graphs, graph databases, or graph-enhanced retrieval to store and query information.

Required methods

def setup(self) -> None

Initialize graph database, embeddings, and any required services.

def reset(self) -> None

Clear the graph. Must remove all nodes and edges.

def teardown(self) -> None

Close connections, stop services, clean up temporary data.

def ingest(self, turns: List[Dict[str, Any]]) -> None

Extract entities and relationships from conversation turns and add to graph.

def recall(self, query: str) -> str

Query the graph for relevant information. May use graph traversal, embedding search, or both.

Required capabilities

-Entity extraction and relationship mapping
-Graph-based or graph-enhanced retrieval
-Knowledge scale handling
-Session isolation (full graph clear on reset)

Applicable benchmarks

Benchmark	Alignment
Knowledge Retrieval	core
Knowledge Scale	core
Truth Arbitration	core
Budget Curves	core
Memory Poisoning	adjacent
Reliability	adjacent

Example implementation

python

"""Example graph memory adapter."""
from typing import Any, Dict, List, Optional
from benchd_harness.adapters.base import BaseAdapter


class MyGraphAdapter(BaseAdapter):

    @property
    def name(self) -> str:
        return "my-graph"

    @property
    def version(self) -> Optional[str]:
        return "1.0.0"

    def setup(self) -> None:
        self._graph = GraphStore(uri="bolt://localhost:7687")

    def reset(self) -> None:
        self._graph.clear_all()

    def teardown(self) -> None:
        self._graph.close()

    def ingest(self, turns: List[Dict[str, Any]]) -> None:
        text = "\n".join(
            f"[{t['role']}]: {t['content']}" for t in turns
        )
        self._graph.extract_and_store(text)

    def recall(self, query: str) -> str:
        nodes = self._graph.query(query, max_hops=2)
        return "\n".join(n.text for n in nodes)

Quick start

Install

bash

pip install benchd-harness

Validate & test