Skip to main content

Solo mode

One self-directed agent that writes its own plan and works it — skipping the planner, plan-review gate, and QA loop. A token-saving path for small jobs. Off by default.

The default AIFactory pipeline runs four roles in separate sessions: planner → coder → QA reviewer → QA fixer. That structure earns its keep on real features — but for a one-line fix or a throwaway script it's overhead: four sessions, several context loads, and a plan-review gate for work that doesn't need one.

Solo mode collapses the pipeline into a single agent. That one agent:

  • writes its own implementation_plan.json,
  • implements the subtasks in the same loop,
  • skips the dedicated planner session, the plan-review gate, and the QA validation loop.

The result is far fewer tokens and sessions for small jobs. The trade-off is no independent QA pass — so it's opt-in and best reserved for low-risk, small-scope tasks. For anything you'd want reviewed, leave it off and use the full pipeline.

Enabling it​

Solo mode resolves through the same channels as other opt-in features, in this order (first match wins):

  1. AIFACTORY_SOLO_MODE environment variable — 1/true/yes/on or 0/false/no/off.
  2. Per-task — the soloMode flag in the task's task_metadata.json, set from the web UI (Task Creation Wizard / settings toggle).
  3. Global — solo.enabled in ~/.aifactory/config.json.
  4. Default — disabled (backward compatible).

The environment variable always wins, so you can force solo mode on (or off) for a whole host regardless of per-task or global settings.

When to use it​

  • Good fit — small, low-risk jobs: a one-file fix, a quick script, a smoke-test spec.
  • Good fit — token-sensitive runs where the full four-session pipeline is overkill.
  • Not a fit — anything you want an independent QA pass on; keep the default pipeline.