OpenRouter Live Multi-Model Gateway & Failover Router

OpenRouter Live Multi-Model Gateway & Failover Router

@openrouter/dsh-live-gateway·v1.6.4·providers

A unified multi-model routing gateway connecting DeepSeek Harness with hundreds of global LLM providers via OpenRouter. Features automated failover, latency routing, and unified billing.

Model-Routing · OpenRouter · Failover · Gateway

GitHubNPM
Installation & Integration
bash
manager:
$ pnpm add @openrouter/dsh-live-gateway

Architecture

OpenRouter Live guarantees high availability and cost efficiency across distributed AI workloads. Relying exclusively on a single LLM vendor leaves mission-critical workflows vulnerable to sudden rate limits (HTTP 429), regional outages, and quota exhausts. This plugin implements a resilient multi-model gateway directly on the Cordis provider layer. Developers specify prioritized fallback cascades—if primary DeepSeek-V3 endpoints throttle, requests seamlessly reroute to verified alternative mirrors without dropping state or failing the active agent turn.

Core Features

01

Hundreds of LLMs via One Key: Route seamlessly across DeepSeek, Claude, Llama, and GPT

02

Sub-Second Failover Cascade: Transparent recovery upon upstream 429 rate limits or timeouts

03

Cost-Optimized Auto-Routing: Matches trivial tasks with ultra-cheap models automatically

04

Live Telemetry & Latency Dashboard: Real-time insights into TTFT latencies and token pricing

Core Workflow

01

Request Routing & Quota Check

Inspects outbound inference payloads, matching requirements with optimal providers.

02

Protocol Normalization

Normalizes prompt, tool calling, and temperature parameters for downstream endpoints.

03

Latency & Availability Probing

Samples live TTFT latencies across provider regions, steering traffic to healthy nodes.

04

Automated Cascade Failover

Transparently triggers secondary fallback models upon HTTP 429/500 errors.

Configuration Parameters Reference (YAML / JSON)

ParameterTypeDefaultDescription
primaryModelstringdeepseek/deepseek-chatPrimary model identifier
fallbackModelsstringdeepseek/deepseek-r1,anthropic/claude-3.5-sonnetOrdered fallback models comma-separated
maxRetryCountnumber3Maximum retry attempts before throwing error