agent-memory

Persistent agent memory is managed through a 3-tier system that stores knowledge across sessions. It organizes information into Shared, Agent-specific, and Conversation-specific tiers, which are automatically injected into the agent's context.

To manage shared project knowledge, use a multi-tiered memory system with a Shared tier for project-wide facts. This allows agents to automatically read and write accumulated architectural decisions or patterns into their active context.

Automatic context injection works by automatically loading saved memory files into the agent's active context window. This reduces manual retrieval steps and ensures the agent has immediate access to relevant historical conversation data.

Yes, you can separate conversation-specific context from agent-specific patterns using a structured memory system. It divides stored knowledge into distinct tiers, allowing individual agents to maintain unique patterns while saving specific dialogue context separately.

No, you do not need to manually retrieve memory files before starting a task. The system automatically injects the relevant persistent memory into the agent's context, although you should write significant findings to the appropriate tier during your task.