AgentDB Memory Patterns

Initialize an AgentDB database with npx agentdb init, then use the createAgentDBAdapter API to store interaction patterns with embeddings. Retrieve relevant context with retrieveWithReasoning using vector similarity and MMR ranking.

Use the session memory pattern: store each message with a sessionId, role, content, and timestamp, then query by sessionId ordered by timestamp. The Skill provides a SessionMemory class template for this workflow.

Yes, AgentDB ships an MCP server that integrates with Claude Code. Run npx agentdb mcp and register it once with claude mcp add agentdb to expose memory operations to the agent.

Yes, the migrateToAgentDB function migrates patterns from a legacy .swarm/memory.db file into AgentDB with validation. It reports how many patterns were migrated and keeps full API backward compatibility.

Large databases result from storing raw embeddings without compression. Enable binary quantization for 32x reduction or scalar for 4x reduction, and run memory consolidation to prune low-scoring patterns.

AgentDB includes nine learning plugins: Decision Transformer, Q-Learning, SARSA, Actor-Critic, Active Learning, Adversarial Training, Curriculum Learning, Federated Learning, and Multi-task Learning. Create them with the create-plugin CLI wizard.