Vibeleaderboard
← All Intel
Intel / article

Deferred Context Engine

Source
factory.ai
Date

Context

Factory's Deferred Engine lets Droid reach many tools without loading every tool's complete instructions on every turn. Each turn starts with a compact index of names, sources and short descriptions; when a task needs a hidden capability, Droid loads its full instructions. Loaded capabilities remain available for the rest of the work. How deferred context works.

In a five-day sample as of its May 2026 post, Factory reported a 15.1% average reduction in estimated input per turn among sessions that triggered -connected tools. The baseline is an estimate of what Droid would have carried had everything been loaded upfront, and the population is those sessions, not all users. A separate Factory approach, compressing conversation history, addresses the same limited prompt space from the other side. Measurement scope.

Terms in this piece · Glossary
  • context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
  • token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
  • MCP — The Model Context Protocol — an open standard that lets any AI assistant plug into any tool or data source without custom integration code.
Recommended reads
Comments

Checking sign-in…

Loading comments…