From 0f2bee833027f38bcbc29c4b0ce36c068497ccf5 Mon Sep 17 00:00:00 2001 From: Nubenetes Bot Date: Sat, 16 May 2026 14:07:39 +0200 Subject: [PATCH] docs: document smart batching and local caching optimizations --- README.md | 2 ++ 1 file changed, 2 insertions(+) diff --git a/README.md b/README.md index 1b231fa5..51e2e18c 100644 --- a/README.md +++ b/README.md @@ -227,6 +227,8 @@ To eliminate configuration overhead and ensure Nubenetes always utilizes the fro 2. **Autonomous Scoring & Ranking**: Models are automatically ranked using a **dynamic regex-based algorithm** that extracts version numbers (e.g., 2.0, 3.1, 4.0). Higher versions are prioritized, ensuring zero-config auto-adoption of future frontier models. Tier bonuses are applied (Ultra > Pro > Flash) to prioritize reasoning depth. 3. **Adaptive Rate Limiting (Exponential Backoff)**: When encountering `429 Too Many Requests` errors, the engine implements an **Exponential Backoff with Jitter** strategy. Instead of immediate rotation, it applies a mandatory wait time that increases with consecutive failures, preventing infinite loops and respecting Google's quota resets. 4. **Concurrency Guard (Semaphore)**: To prevent saturating API quotas during high-volume operations (like V2 inventory enrichment), the system utilizes an **Asyncio Semaphore**. This restricts the number of concurrent AI calls (e.g., max 5), ensuring a steady, reliable flow that stays within RPM (Requests Per Minute) limits. +5. **Smart AI Batching (90% Traffic Reduction)**: Instead of processing one link per call, the system groups up to **10 resources into a single AI prompt**. This strategic packaging reduces total API calls by 90%, drastically lowering the risk of `429` errors while optimizing token density for Identity A. +6. **Pre-Flight Local Caching**: The engine performs an autonomous look-up in `data/inventory.yaml` before any AI operation. If a resource is already indexed and described, it is skipped in the enrichment phase. This makes the marginal cost of repository maintenance near-zero. ### AI Intelligence & Observability (Transparency) As of May 2026, Nubenetes implements a **Total Transparency Protocol** for AI operations. Every curation cycle is tracked to ensure maintainers understand the cost, quality, and infrastructure behind the agentic decisions: