Agent Workforce

Agent Workforce lets one local Bestie runtime keep a set of fixed agent profiles and a local task queue. It is useful when recurring work benefits from a named role and an inspectable task record instead of an ad hoc chat request.

Agent profiles and tasks stay in the local Bestie runtime. Creating or assigning work does not start it automatically: use bestie agents run when you are ready to process queued tasks.

Create an agent

Create a profile with a stable ID, a display name, a role, and an operating description:

bestie agents hire \
  --id docs \
  --name "Docs Editor" \
  --role documentation \
  --description "Maintain accurate local product documentation." \
  --model openai/gpt-4o-mini \
  --tools internal.read_url,internal.write_file

--model and --tools are optional. The command writes a prompt file for the agent and prints its path. Inspect configured agents with:

bestie agents list
bestie agents show docs

Assign and run work

Assign a task to an existing agent:

bestie agents assign \
  --agent docs \
  --title "Review onboarding guide" \
  --brief "Check the onboarding guide against the current CLI behavior."

Review the queue, then run queued work:

bestie agents tasks
bestie agents tasks --agent docs --status queued
bestie agents run
bestie agents run --agent docs --limit 1

Use --watch to keep processing newly queued tasks until you stop the command. --interval-ms controls the polling interval and must be a positive integer.

bestie agents run --watch --interval-ms 5000

Task statuses are queued, in_progress, done, blocked, and canceled. You can inspect a task or record a status and result manually:

bestie agents task <id>
bestie agents task-status <id> --status blocked --result "Waiting for the provider configuration."

Pause, resume, and remove

Pause an agent to stop it from processing queued tasks, then resume it when ready:

bestie agents pause docs
bestie agents resume docs

To remove a profile:

bestie agents remove docs

Review tasks before removal. Keep task briefs free of secrets because they are stored locally with the workforce task record.