Update from code changes: document QiForge oracle latency knobs

[mintlify]

Summary

Document the new MAIN_REASONING_EFFORT env var and the always-on request-path caching + memory-context compaction that ship with the QiForge oracle latency work.

Changes

• build-an-oracle/reference/environment-variables.mdx — new Latency tuning section covering MAIN_REASONING_EFFORT (low/medium/high, default medium, main role only) plus a note describing the always-on request-path caches (per-user SqliteSaver reuse, history-skipping build-time checkpoint read, room-keyed userContext cache) and the always-on memory-context compaction (dedup across the six memory buckets, per-entity summary and whole-block caps, recall-footer overflow).
• build-an-oracle/reference/bundled-plugins/memory.mdx — updated the memory-context fetcher description: the 5-minute cache is now keyed by roomId (not sessionId), so a fresh session for the same room reuses cached context. Added a paragraph describing the dedup + budget applied to the prompt block.

Context

Source PR: ixoworld/qiforge#216 — added memory compaction, request-path caches, and reasoning-effort tuning. The cache flags and compaction flag were promoted to always-on in the same PR; only MAIN_REASONING_EFFORT remains as an operator-facing env var.

[mintlify]

Preview deployment for your docs. Learn more about Mintlify Previews.

Project	Status	Preview	Updated (UTC)
ixoworld	🟢 Ready	View Preview	Jul 7, 2026, 1:20 AM