Beta self-hosted HTTP and LLM proxy with Redis-backed caching, non-streaming LLM idempotency, and budget guardrails.
Drop-in idempotency layer for POST endpoints: atomic Redis claiming, byte-identical replay, in-flight collapsing. Proof asserted in tests: 50 concurrent duplicate payment requests, exactly one charge. Includes the crash-window test middleware demos hide, closed with a DB uniqueness constraint (ADR explains which layer owns which guarantee).
ComtradeBench OpenEnv: execution-grounded benchmark for reliable LLM tool-use under adversarial trade-data API conditions.
Model-serving gateway with dynamic micro-batching (fill-to-32 or 4ms deadline) and load shedding: +54% throughput and p99 cut from 2,289ms to 915ms at 64-way concurrency vs naive serving. Documents, with measurements, when batching is pure loss.
Validating, idempotent ingestion gateway for FHIR-shaped healthcare data: cross-field validation at the door (claim totals must equal item sums), Idempotency-Key replay/409 semantics, batch vs transaction bundles, and a requeueable dead-letter queue. ~600 resources/s measured; bad data never 500s.
可复现的 AI API 中转站测速工具：测试 OpenAI 兼容接口成功率、P50/P95 延迟、错误率和稳定性。
Cost-Aware API Reliability RL Environment using OpenEnv
Add a description, image, and links to the api-reliability topic page so that developers can more easily learn about it.
To associate your repository with the api-reliability topic, visit your repo's landing page and select "manage topics."