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

Description and use

DHP provides durable execution for AI agents, automatically recovering from a worker kill within ~7 s without data loss or rework.

Problem

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

Outcome

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.

Workflow

• 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.

Evidence

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

Source

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

Source

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

Source