kubeswarm Vector Stores - Qdrant, Pinecone, Weaviate

kubeswarm supports vector store backends for persistent agent memory on Kubernetes via the SwarmMemory CRD. Store and retrieve embeddings across tasks.

Supported Backends

Provider	Endpoint format	Use case
Qdrant	qdrant://host:6334/collection	Self-hosted, Kubernetes-native
Pinecone	pinecone://index-name	Managed cloud service
Weaviate	weaviate://host:8080/class	Self-hosted or cloud

Configuration

apiVersion: kubeswarm.io/v1alpha1
kind: SwarmMemory
metadata:
  name: agent-memory
spec:
  backend: vector-store
  vectorStore:
    provider: qdrant
    endpoint: qdrant.default.svc.cluster.local:6334
    collection: research-findings
    ttlSeconds: 0                  # 0 = no expiry
  embedding:
    model: text-embedding-3-small
    provider: openai
    dimensions: 512

Embedding Providers

Provider	Models	Notes
OpenAI	text-embedding-3-small, text-embedding-3-large	Requires OPENAI_API_KEY
Google	text-embedding-004	Requires GOOGLE_API_KEY
Voyage AI	voyage-3, voyage-3-lite	Requires VOYAGE_API_KEY

Usage with Loop Policy

Vector memory is used during the agent's tool-use loop via spec.runtime.loop.memory:

spec:
  runtime:
    loop:
      memory:
        ref:
          name: agent-memory       # SwarmMemory reference
        store: true                # write summaries after tool calls
        retrieve: true             # fetch similar findings before tool calls
        topK: 5
        minSimilarity: 0.70

See Loop Policy for the full deep-research runtime configuration.