async-job-orchestration

To manage long-running async LLM jobs over 2 minutes, you can dispatch non-blocking tasks using the `claude_request_async` command, which handles parallel execution and tracks job states efficiently. This ensures durable results that survive gateway restarts.

Cache awareness in async LLM job orchestration utilizes cache-aware `promptParts` and `cache-state://` resources to optimize caching. This mechanism reduces redundant processing during parallel executions and maintains efficient prompt state tracking.

To track job state for parallel LLM executions across gateway restarts, the orchestration system provides durable job state tracking. This feature monitors job lifecycles, handles polling, and ensures results persist without data loss during interruptions.

Yes, you can run multiple non-blocking LLM analysis tasks in parallel. The orchestration system supports parallel execution of asynchronous jobs, making it ideal for implementing review workflows and long-running analysis tasks simultaneously.

You should use asynchronous execution for large language model tasks when operations exceed 2 minutes, require parallel processing, or need non-blocking execution. This approach is ideal for long-running analysis and review workflows requiring durable results.