I Killed My AI Agent 14 Times. It Kept Working.

说明与用途

DHP 为 AI 代理提供持久化执行，在约 7 秒内自动从被杀死的工作进程恢复，无数据丢失或重复工作。

问题

Agent frameworks lack durability; when a worker dies during execution, the run fails and must restart, causing lost progress and wasted resources.

结果

DHP enables sub‑second (≈7 s) kill‑to‑resume for LLM workloads, achieving zero rework across 18 randomized kills (14 workers, 4 supervisors) and 1,600 checkpoints, outperforming Temporal’s ~12 s stateful failover.

工作流

• 1. Worker starts task and begins processing.
• 2. Worker is killed (kill -9) mid-execution.
• 3. Supervisor detects missed heartbeats after ~6.5s and marks handoff orphaned.
• 4. Standby worker claims the handoff, reads last checkpoint via TCP.
• 5. Standby resumes from next step and completes the task.

证据

I surveyed the landscape — LangGraph, CrewAI, AutoGen, the OpenAI Agents SDK. The answer is the same everywhere: nothing. No death detection. No scheduled resume.

来源

Result: 60/60 pages fetched. Zero refetched. The kill was invisible in the output — just a 6.5-second pause.

来源

I proved this by destroying an entire host — disk and all — mid-run.

来源