tiered-memory

Reduce context payload size for Squad agents by implementing a three-tier memory model that separates data into Hot, Cold, and Wiki tiers, achieving 20-55% context reduction per spawn. This eliminates unnecessary old noise from full history loading.

Tiered memory management improves performance by supporting on-demand loading of Cold and Wiki memory tiers, which prevents loading full context history for every spawn and significantly reduces computational overhead.

Yes, you need the Scribe dependency to act as a memory coordinator for managing Hot, Cold, and Wiki memory tiers, which is required to enable on-demand loading and reduce payload sizes.

Loading full context history increases computational overhead because it results in large payloads and unnecessary old noise for every spawn, which a three-tier memory model solves by shrinking payload size by 20-55%.

The best way to optimize context for complex Squad agent tasks is enabling a three-tier memory model that achieves 20-55% context reduction per spawn, significantly lowering the computational load while handling large context histories.