AxContextPolicyPreset

type AxContextPolicyPreset = "full" | "adaptive" | "lean" | "checkpointed";

Defined in: https://github.com/ax-llm/ax/blob/5b28f9093bb70863b59459bb6df5062d005bce41/src/ax/agent/rlm.ts#L436

Opinionated context replay presets for the Actor loop.

• full: Keep prior actions fully replayed with minimal compression. Best for debugging or short tasks where the actor should reread exact old code/output.
• adaptive: Keep live runtime state visible, preserve recent or dependency-relevant actions in full, keep discovery docs available by default, and collapse older successful work into checkpoint summaries as context grows. Reliability-first defaults favor summaries before deletion. Best default for long multi-turn tasks.
• lean: Most aggressive compression. Keep live runtime state visible, checkpoint older successful work, and summarize replay-pruned successful turns instead of replaying their full code blocks. Reliability-first defaults still preserve recent evidence before deleting older low-value steps. Best when character-based prompt pressure matters more than raw replay detail.
• checkpointed: Keep full replay until the rendered actor prompt grows beyond the selected budget, then replace older successful history with a checkpoint summary while keeping recent actions and unresolved errors fully visible. Best when you want conservative, debugging-friendly replay until prompt pressure becomes real.