CHAPTER 05
Continuity
English summary
DeerFlow separates execution recovery, conversational continuity, stream reconnection, and auditability. Checkpoints, summaries, long-term memory, stream buffers, and run events have distinct identities, durability, timing, and failure semantics.
01Execution continuity through ThreadState and checkpoints
02Conversation continuity through summaries, durable context, and memory
03Reconnect and gap recovery through StreamBridge
04Audit boundaries through RunEventStore and RunJournal
Read the implementation
This edition deliberately stops at a concise overview. The links below open the exact DeerFlow source files used by the complete Chinese chapter.
agents/thread_state.py · idempotent sandbox identity reducerruntime/checkpoint_state.py · mode-aware state accesssummarization_middleware.py · summarize before replacementdynamic_context_middleware.py · one agent-entry snapshotmemory_middleware.py · async manager boundarydeermem/core/queue.py · selective backpressurefrontend/core/api/api-client.ts · reload durable state on a replay gapruntime/journal.py · restore failed batches before raising