Problem or Use Case

The orchestrator queues up all ready tasks and moves them to In Progress. When running with a local hosted LLM this can cause the tasks to never be completed and even respawned multiple times, making it difficult or impossible for the task to be marked complete.

This requires me to actively watch the tasks in progress and complete lists so that I do not overwhelm the agent. I must then manually move the other tasks to ready when the queue is empty.

Respawning a task will also overwrite any progress that was currently made forcing the new agent to start the task from scratch.

Proposed Solution

Add a configuration setting to the config.yaml file under the kanban section.

kanban
  active_task_limit: 2
  # or the same as for the cron jobs
  max_parallel_jobs: 2

And then the In Progress field is checked across all boards before any new tasks are moved from Ready to In Progress. It should also check the current active tasks before spawning a new agent to process a task that is already actively being run.

Alternatives Considered

Using a local hosted LLM obviously it may be slower to complete larger tasks than a paid host. The delay in processing should not interfere with the correct process to complete a task though.

I have already adjusted the dispatch interval to 3 minutes from every 1 minute, this does not seem to fix the issue.

I have also adjusted the cron max_parallel_jobs to 2, unrelated but easier for my local hosted LLM.

Feature Type

Other

Scope

Small (single file, < 50 lines)

Contribution

• I'd like to implement this myself and submit a PR

Debug Report (optional)