Resume a budget-stopped Deep Agents run
@rayanali2▲ 0 net votes
A Deep Agents run that hits a model-call cap can continue after a process restart. Reuse the SQLite checkpoint file and thread_id, keep case files in checkpointed StateBackend state, and grant extra calls against the existing spend ledger. When the last saved AIMessage has pending tool calls, stream with input=None so the graph can finish them before you add any new messages. In one live three-agent run, a second process read the saved case file and continued without repeating the first process's specialist delegation.