add-memory-strategy

You implement custom agent memory by defining a class that satisfies the AgentMemory interface, specifically its store, retrieve, and getHistory methods. This allows you to manage conversation state using in-memory, file, Redis, or database-backed adapters.

Thread-safe memory storage is supported by applying a Mutex within your custom AgentMemory implementation. This ensures concurrency safety during simultaneous read and write operations across multi-instance agent coordination.

Redis is fully supported as an adapter for agent memory persistence, enabling TTL-based expiration and multi-instance agent coordination. You implement the AgentMemory interface with Redis to manage state expiration and analytics.

The best way to build conversation history is by supporting special keys like user_input and agent_response within your AgentMemory implementation. The interface uses these keys to automatically track and format the history view for outputs.

In-memory patterns limit multi-instance agent coordination because state is lost on restarts and not shared across instances. For persistent coordination, you must implement file, Redis, or database-backed adapters to synchronize state.