CONTINUOUS SHIPMENT

AgentMink Changelog

Weekly product releases, new foundation model integrations, sub-millisecond edge latency optimizations, and developer infrastructure features.

v1.4.0 September 2, 2026
MODEL SUPPORT PERFORMANCE NEW FEATURE

DeepSeek-V3 Integration, 45% Faster Pre-Dispatch & Sub-50ms Failover

DeepSeek-V3 & DeepSeek-R1 BYOK Support: Added full BYOK routing for DeepSeek chat and reasoning models with zero token markup.
Concurrent Pre-Dispatch Pipeline: Parallelized KV authentication record lookup, quota lock verification, daily spend checking, and JSON body parsing via Promise.all, cutting proxy overhead by 45% (down to <1.2ms).
Sub-50ms Zero-Downtime Failover: Intercepts upstream HTTP 429 and 500 crashes instantly at the edge, rerouting to pre-warmed secondary model sockets before client timeouts. 
// Direct DeepSeek dispatch via AgentMink
const response = await fetch("https://api.agentmink.com/v1/chat/completions", {
  method: "POST",
  headers: {
    "Authorization": "Bearer mk_live_...",
    "X-AgentMink-Strategy": "cost_first"
  },
  body: JSON.stringify({
    model: "deepseek/deepseek-chat",
    messages: [{ role: "user", content: "Summarize API schema." }]
  })
});
v1.3.5 August 24, 2026
MODEL SUPPORT NEW FEATURE

Google Gemini 2.0 Flash Support & Real-Time Spending Circuit Breakers

Google Gemini 2.0 Flash & Pro: Direct streaming passthrough and multimodal token ledgering for Gemini 2.0 endpoints.
Sub-Millisecond Spend Caps: Configurable daily spending limits with instant HTTP 402 circuit breakers to prevent runaway agent loops from draining cloud budgets.
Slack Incident Webhooks: Immediate automated dispatch when upstream provider rate limits or quota triggers are tripped.
v1.3.0 August 10, 2026
SECURITY & BYOK PERFORMANCE

AES-256-GCM BYOK Vault & 6-Hour Distributed Pricing Cache

Authenticated AES-256-GCM Encryption: Vendor API keys are stored encrypted at rest and decrypted strictly in volatile RAM at the edge during upstream dispatch.
In-Process Pricing Cache: Global model pricing matrices are cached in-memory across edge isolates for 6 hours, eliminating cold KV read latency on the hot request path.
Pure BYOK Zero-Markup Standard: Verified 100% pass-through pricing model\u2014developers pay direct vendor tiers with $0 platform markup.
v1.2.0 July 22, 2026
NEW FEATURE

1-Line OpenAI & Vercel AI SDK Drop-In Compatibility

Universal OpenAI SDK Compatibility: Developers can route all inference by changing just 1 line: baseURL: "https://api.agentmink.com/v1".
Dynamic Routing Strategy Headers: Select between cost_first, quality_first, and latency_first per-request via HTTP headers.
Vercel AI SDK Provider Wrapper: First-class support for streaming UI completions with next-gen AI frameworks.
v1.1.0 July 05, 2026
NEW FEATURE SECURITY

Non-Blocking Telemetry Ledgering & GitHub OAuth Auth

Asynchronous Telemetry: Multi-provider token consumption, execution costs, and latency metrics ledgered asynchronously using Cloudflare waitUntil() with zero TTFT penalty.
GitHub & Google OAuth2: 1-click developer onboarding with automatic Developer Trial key generation.
Synchronous SWR Cache Hydration: Zero-flash dashboard loading across workspace views.
v1.0.0 June 18, 2026
MAJOR RELEASE 300+ POPS

AgentMink 1.0 General Availability

Global Edge Network: Deployed across 300+ Anycast edge data centers worldwide.
Stateless Ephemeral V8 Isolates: Zero prompt retention, zero disk storage, meeting strict GDPR Article 17 and CCPA enterprise standards.
Enterprise 99.99% SLA: Mission-critical infrastructure for high-throughput AI platforms.