Nenya

v0.8.0 Feature

This release adds 4 notable features for engineering teams evaluating rollout.

Published 13d LLM Frameworks

View tool

✓ No known CVEs patched

Read the diff → Tool health → What is this tool? →

✓ No known CVEs patched in this version

Topics

ai ai-gateway ai-governance ai-proxy ai-safety ai-security

+14 more

ai-tools anthropic deepseek litellm-alternative llm llm-gateway observability ollama openai openai-compatible-api openai-proxy opencode-zen proxy zhipu-ai

Summary

AI summary

Updates affect Anthropic, cache injection, metrics tracking, routing stripping, and proxy resilience patterns.

Changes in this release

Type	Severity	Summary	CVE
Feature
Feature	Low	Adds adaptive thinking support for newer Anthropic models. Adds adaptive thinking support for newer Anthropic models. Source: llm_adapter@2026-07-14 Confidence: high	—
Feature	Low	Injects prompt_cache_key for xAI/OpenAI providers in cache layer. Injects prompt_cache_key for xAI/OpenAI providers in cache layer. Source: llm_adapter@2026-07-14 Confidence: high	—
Feature	Low	Tracks reasoning token usage in streaming metrics path. Tracks reasoning token usage in streaming metrics path. Source: llm_adapter@2026-07-14 Confidence: high	—
Feature	Low	Injects reasoning_effort for Grok models in xAI provider. Injects reasoning_effort for Grok models in xAI provider. Source: llm_adapter@2026-07-14 Confidence: high	—
Feature	Low	Strips non‑standard bare reasoning field from routing layer. Strips non‑standard bare reasoning field from routing layer. Source: llm_adapter@2026-07-14 Confidence: high	—
Bugfix
Bugfix	Medium	Fixes context‑overflow pattern handling and adds HTTP 529 retry logic. Fixes context‑overflow pattern handling and adds HTTP 529 retry logic. Source: llm_adapter@2026-07-14 Confidence: high	—
Bugfix	Medium	Restores missing context‑limit error patterns in resilience module. Restores missing context‑limit error patterns in resilience module. Source: llm_adapter@2026-07-14 Confidence: high	—
Bugfix	Medium	Addresses reasoning token tracking bugs found during review. Addresses reasoning token tracking bugs found during review. Source: llm_adapter@2026-07-14 Confidence: high	—
Bugfix	Low	Adds hex validation for OpenAI cache key in review tests. Adds hex validation for OpenAI cache key in review tests. Source: llm_adapter@2026-07-14 Confidence: high	—
Refactor	Low	Aligns constant declarations with gofmt formatting standards. Aligns constant declarations with gofmt formatting standards. Source: llm_adapter@2026-07-14 Confidence: high	—

Full changelog

Changelog

c29e77b99a243e2ac6554f627ecee6ef433adb77 Merge phase-010: Context-overflow patterns + HTTP 529 retry
6de207a3d24d120716f50a22c0ccebeb4455c38c Merge phase-011: Model version detection utility
1d125415d5440b5293feb623d1d0f572429e9126 Merge phase-012: Adaptive thinking for Anthropic
a34ed2df1acafd68dd5082b32f0a5e7bb7c8f075 Merge phase-013: xAI reasoning_effort injection
57370ab4587389754af8be798e34f276534b5b52 Merge phase-014: Strip bare reasoning field
bb5b9f2933d511d32c698531118b90214e0c4557 Merge phase-015: Prompt cache key injection
4e3093b74aae63137e293c68b7f04ba3285fd0a5 Merge phase-016: Reasoning token usage tracking
e6988517969d51fd75b033e17200453671ffe856 feat(anthropic): add adaptive thinking support for newer Anthropic models
776c93933697dcb70d510a9b524b944cc1fc8719 feat(cache): inject prompt_cache_key for xai/openai providers
59802cffd57a152c375d186337fe5c81d5aea436 feat(metrics): add reasoning token tracking to streaming path
371f22b1b57a943329a63be5bfcc842ed919d591 feat(routing): strip non-standard bare reasoning field
bed99620c8d307d1314faf545dbfb46e940ad226 feat(util): add Anthropic model version detection utility
535873f4b68446721aed08568b98ca4c294ef7eb feat(xai): inject reasoning_effort for reasoning-capable Grok models
f3a636f39a788f81d0fdb17c5770303fdca0ebb4 fix(format): align constant declarations with gofmt
584e84679807c23f581c58cf0861b2197e7d9a44 fix(metrics): address reasoning token tracking bugs found in review
b87c69ff5f1c3d865e93f2bae1c53c5c63166fa4 fix(proxy): expand context-overflow patterns and add HTTP 529 retry
a04575834836da6fcfd7c14e16d89eb4227aad63 fix(resilience): restore missing context-limit error patterns
811288c494da6a86c1820f4dce0e851299312df2 fix(review): add hex validation for OpenAI cache key test
a212b5de5189891b109f2ff5c8a6f06b7ffa5584 fix(review): address Phase 012 review findings
dec422534c868aafacde667ca681753b051ae7d1 fix(review): address Phase 013 review findings
21a50ff62dc6c5277bd3e98db7eca1d4f9abe3b6 fix(review): address Phase 014 review findings