让 GPT-6 智能体扛住崩溃重启的 12 个开源仓库

virgilxbt @virgilxbt原推发布于 2026.10.13

在 X 上打开原推

AI agents are getting f…cking impossible to stop. 12 GitHub repos that let a GPT 6 system outlive crashes, restarts, failed tools and lost sessions. 01 Restate > https://github.com/restatedev/restate -> Durable execution for agents and workflows. MAKE EXECUTION SURVIVE 02 DBOS > https://github.com/dbos-inc/dbos-transact-py -> Python workflows that resume after a cut. 03 Windmill > https://github.com/windmill-labs/windmill -> Scripts, workflows, background jobs. 04 http://Trigger.dev > https://github.com/triggerdotdev/trigger.dev -> Durable tasks with retries. KEEP THE WORKERS MOVING 05 NATS > https://github.com/nats-io/nats-server -> Messaging for distributed workers. 06 Celery > https://github.com/celery/celery -> Tasks spread over worker processes. 07 BullMQ > https://github.com/taskforcesh/bullmq -> Redis job queues, retries included. SAVE THE STATE 08 Valkey > https://github.com/valkey-io/valkey -> Fast state and coordination. 09 PostgreSQL > https://github.com/postgres/postgres -> Jobs, checkpoints, execution history. 10 MinIO > https://github.com/minio/minio -> Artifacts and intermediate outputs. RECOVER AND VERIFY 11 OpenTelemetry > https://github.com/open-telemetry/opentelemetry-collector -> Traces and telemetry across services. 12 Argo Workflows > https://github.com/argoproj/argo-workflows -> Container workflows on Kubernetes. The architecture: Task -> queue -> execute -> checkpoint -> failure -> restore -> retry -> verify -> continue. 3 builds to bookmark: > Python agents: DBOS -> Celery -> PostgreSQL. > Distributed workers: Restate -> NATS -> Valkey. > Production workflows: Windmill -> Argo Workflows -> OpenTelemetry. The big change is not GPT 6 Astra doing more. It is everything around it surviving when a task breaks. A crashed process is not a lost mission. The next generation of agents will not need perfect runs. They will need systems that know how to continue ↓

这条是引用着另一条写的：「工作室用Grok Bot例行任务每天自动排发视频」