| Agent Memory | DB-backed agent memory using DocumentStore + IndexStore. Reads agent context from flow headers (agent_id). Provides context-window management with compression-on-retrieval, JSONL session logging, and transcript finalization. Compressed messages include a source reference to the raw JSONL log. |
| Buffer Memory | Attach this node to an Agent or LLM to keep track of the conversation history. |
| Compact Memory | Dynamic memory with compression that automatically manages context window limits by compressing old messages while preserving important information. |
| Dynamic Memory | Attach this node to an Agent or LLM to keep track of the conversation history. |
| Skills Memory | File-based memory for Skills that writes messages directly to the workspace container. Files are instantly available to the agent and synced to GCS in the background. |
| Tiered Memory | Three-tiered memory: No Compression → Tool Compression → Aggressive Tool & Response Compression |
| Workspace Memory | File-based memory for Workspaces that writes messages directly to the workspace container. Files are instantly available to the agent and synced to GCS in the background. Pending writes are stored in session state for durability. |
| Workspace Memory | File-based memory for Workspaces that writes messages directly to the workspace container. Files are instantly available to the agent and synced to GCS in the background. Pending writes are stored in session state for durability. |
| Workspace Memory 2 | Enhanced memory with hybrid BM25 + vector search. Stores messages in JSONL files with async embedding generation. Provides search_memory, save_to_memory, compact_memory, and other MCP tools. Uses /workspace/.memory/ (separate from Workspace Memory's .session/). |