agents-features-memory

Provides the ChatMemory feature that persists and restores conversation history between agent runs.

Overview

The agents-features-memory module enables AI agents to maintain continuity across multiple sessions by storing and loading the full conversation history. When a new agent run starts, the previous messages are automatically injected into the prompt; when the run completes, the updated history is saved back to the configured provider.

Main Components

• ChatMemory — the installable agent feature. Intercepts strategy start and completion events to load and store conversation history via a ChatHistoryProvider.

• ChatMemoryConfig — configuration holder. Accepts a ChatHistoryProvider implementation and an optional list of ChatMemoryPreProcessors applied to the history before it is used or saved.

• ChatHistoryProvider — interface for reading and writing conversation history, keyed by a conversation (run) identifier. Implementations can target any storage backend (in-memory, SQL, file system, etc.).

• InMemoryChatHistoryProvider — built-in, thread-safe in-memory implementation of ChatHistoryProvider, useful for testing and short-lived sessions.

• ChatMemoryPreProcessor — interface for transforming the message list before it is injected into the prompt or persisted. The module ships with a sliding-window pre-processor that trims history to a configurable maximum number of messages.

Goals

• Provide conversation continuity across agent runs without requiring changes to agent strategy logic.

• Keep the storage backend pluggable so teams can use any persistence layer.

• Allow history shaping (e.g. truncation, filtering) through a composable pre-processor pipeline.