Synaptic Memory

await graph.add("Payment Outage Postmortem", "PG API timeout caused...")

# LLM auto-generates:
# kind: LESSON
# tags: ["payment", "PG", "timeout", "circuit-breaker"]
# search_keywords: ["payment failure cause", "PG outage response", "API timeout fix"]
# search_scenarios: ["searching past cases when payment system fails"]
# relations to existing nodes: --[LEARNED_FROM]--> "deployment decision"

from synaptic import SynapticGraph, ActivityTracker

async def main():
    graph = SynapticGraph.memory()
    tracker = ActivityTracker(graph)

    # Search past experiences (intent auto-inferred)
    result = await graph.agent_search("DB migration failure")

    # Record a decision
    session = await tracker.start_session(agent_id="my-agent")
    decision = await tracker.record_decision(
        session.id,
        title="Choose PostgreSQL",
        rationale="Need vector search + ACID",
        alternatives=["MongoDB", "SQLite"],
    )

    # Record outcome → auto Hebbian learning
    await tracker.record_outcome(
        decision.id,
        title="Migration succeeded",
        content="Achieved zero downtime",
        success=True,
    )

from synaptic import SynapticGraph

graph = SynapticGraph.sqlite("knowledge.db")
await graph.backend.connect()

# RuleBasedClassifier + RelationDetector + Ontology included automatically.
# Just add content — kind and relations are auto-classified.
await graph.add("Refund Policy", "Refunds available within 7 days...")  # → kind=RULE (auto)

from synaptic import SynapticGraph
from synaptic.backends.sqlite import SQLiteBackend
from synaptic.extensions.llm_provider import OllamaLLMProvider

graph = SynapticGraph.full(
    SQLiteBackend("knowledge.db"),
    llm=OllamaLLMProvider(model="qwen3:0.6b"),
    embed_api_base="http://localhost:8080/v1",
    embed_model="BAAI/bge-m3",
)
await graph.backend.connect()

# LLM auto-generates: kind classification + tags + search keywords + search scenarios
# Embeddings include search_keywords → improved vector search accuracy
# Semantic relations auto-detected against existing nodes (DEPENDS_ON, LEARNED_FROM, etc.)
node = await graph.add("Payment Outage Postmortem", "PG API timeout caused...")

from synaptic import SynapticGraph, OpenAIEmbeddingProvider
from synaptic.backends.sqlite import SQLiteBackend

graph = SynapticGraph(
    SQLiteBackend("knowledge.db"),
    embedder=OpenAIEmbeddingProvider("http://gpu-server:8080/v1", model="BAAI/bge-m3"),
)
await graph.backend.connect()

# Auto: title + content → vector generation → stored
await graph.add("Deployment Strategy", "Blue-green deployment for zero downtime")
# Auto: query → vector generation → FTS + vector hybrid search
result = await graph.search("deployment approach")

from synaptic import SynapticGraph

graph = SynapticGraph.kuzu("knowledge.kuzu")
await graph.backend.connect()
await graph.add("Deploy Policy", "Auto-deploy after PR merge")

from synaptic import SynapticGraph
from synaptic.backends.composite import CompositeBackend
from synaptic.backends.kuzu import KuzuBackend
from synaptic.backends.qdrant import QdrantBackend
from synaptic.backends.minio_store import MinIOBackend

composite = CompositeBackend(
    graph=KuzuBackend("knowledge.kuzu"),
    vector=QdrantBackend("http://localhost:6333"),
    blob=MinIOBackend("localhost:9000", access_key="minio", secret_key="secret"),
)
await composite.connect()

graph = SynapticGraph.full(composite, embed_api_base="http://gpu-server:8080/v1")

# Internal routing:
# - embedding → Qdrant, content > 100KB → MinIO, everything else → Kuzu

from synaptic import OntologyRegistry, TypeDef, PropertyDef, build_agent_ontology

ontology = build_agent_ontology()

# Add custom type
ontology.register_type(TypeDef(
    name="incident",
    parent="agent_activity",
    description="Production incident",
    properties=[
        PropertyDef(name="severity", value_type="str", required=True),
    ],
))

graph = SynapticGraph(backend, ontology=ontology)
# → Auto-validated on graph.add() and graph.link()