PostgreSQL Memory

HTTP memory service backed by PostgreSQL for persistent conversation storage.

Installation


make postgres-memory-install  # Installs PGO and deploys the service

Database Access


# Port forward to PostgreSQL
kubectl port-forward \
  $(kubectl get pod -l postgres-operator.crunchydata.com/cluster=postgres-memory-postgres,postgres-operator.crunchydata.com/role=master -o name) \
  5432:5432
 
# Get password and connect
export PGPASSWORD=$(kubectl get secret postgres-memory-postgres-pguser-postgres \
  -o jsonpath='{.data.password}' | base64 -d)
psql -h localhost -U postgres -d memory
 
# Or just
make postgres-connect

Local Development


make postgres-memory-dev

Configuration

The PostgreSQL Memory service stores conversation history and agent memory across sessions. It integrates with ARK agents to provide persistent memory capabilities.

Environment Variables

DATABASE_URL - PostgreSQL connection string (auto-configured in cluster)

Usage

Once installed, agents can reference the memory service in their configuration to enable persistent conversation storage.

This example shows how to use a memory object to track occurrences of a keyword (bazinga) and terminate the conversation if it is repeated more than three times.


 
apiVersion: ark.mckinsey.com/v1alpha1
kind: Agent
metadata:
  name: agent-rr
spec:
  prompt: Append "Bazinga!" to the user input and say it. If you see Bazinga 3 times, end the conversation.
  tools:
    - type: built-in
      name: terminate
---
apiVersion: ark.mckinsey.com/v1alpha1
kind: Team
metadata:
  name: team-rr
spec:
  members:
    - name: agent-rr
      type: agent
  strategy: "round-robin"
---
apiVersion: ark.mckinsey.com/v1alpha1
kind: Query
metadata:
  name: query-rr
spec:
  input: "Sheldon says:"
  memory:
    name: postgres-memory
  targets:
    - type: team
      name: team-rr

You try this out by :


kubectl apply -f samples/teams/round-robin-with-terminate.yaml


fark team team-rr query-rr

Try this query more than three times to test it.