// How It Works

How Nyquest
Optimizes Your Requests

Every message sent through Nyquest flows through a multi-stage optimization pipeline. It normalizes, optimizes, and routes your prompts before they reach any AI model — improving response quality while reducing latency.

// Pipeline

Six-Stage Optimization Pipeline

From raw prompt to optimized dispatch in milliseconds.

01 NORMALIZE

Prompt Normalization

Extract constraints, detect conflicts, remove redundant instructions, preserve format requirements. Reduces entropy before compression.
02 AGENT MODE

Agent Optimization

7-strategy pipeline for agentic systems: tool result pruning, schema minimization, thought compression, error dedup, sliding window, cache injection.
03 CACHE REORDER

Prefix Caching

Sorts tool definitions and system blocks into deterministic order for maximum provider-side prefix cache hits. Transparent to the model.
04 COMPRESS

Rule Compression

350+ compiled rules across 18 categories. Filler removal, verbose substitution, code minification, format optimization. Sub-2ms latency.
05 SEMANTIC LLM

Neural Condensation

Semantic engine intelligently condenses system prompts and conversation history. Fires on large messages only. Auto-fallback on timeout.
06 AUTO-SCALE

Dynamic Level + Forward

Dynamically adjusts compression based on context window utilization. Small prompts get lighter compression; large contexts ramp to max. Then forwards to provider.
// Optimization

Adaptive Request Optimization

Nyquest dynamically adjusts optimization intensity based on prompt size and context window utilization. Small prompts get lighter treatment; large contexts get full-pipeline optimization. Fidelity always comes first.

Text Content
System prompts, user messages, and conversation history. Filler phrases, verbose phrasing, redundant instructions, and AI-generated noise are cleaned automatically.
Code & Data
Code blocks, JSON payloads, and structured data are optimized for efficient model consumption while preserving semantics.
Never Modified
Tool/function schemas, image blocks, audio blocks, API response bodies, and cache control markers pass through untouched. Zero schema corruption.
// Benchmarks

v3.1.1 Production Benchmarks

Production benchmarks on Linux. Optimized backend. April 2026.

TestResultDetail
Health Check✅ Passv3.1.1 · optimization engine
Health Throughput549 req/sp50: 1.82ms · p99: 2.25ms
Concurrent (20 workers)980 req/sp50: 13.6ms · p99: 35.2ms
Live Proxy✅ PassNegative overhead (−161ms vs direct)
SSE Streaming✅ Pass349ms TTFB
OpenAI Endpoint✅ PassCorrect chat.completion format
Resource Usage71.4 MB0.0% system memory · 1.2% CPU
// Architecture

High-Performance Backend

Nyquest v3.1.1 is a purpose-built backend. HTTP server, optimization engine, provider routing, streaming relay — all in one lightweight package.

Performance

980 req/s concurrent throughput
1.82ms p50 latency
71.4 MB memory footprint
Lightweight package size

Dual API Surface

/v1/messages — Anthropic Messages API
/v1/chat/completions — OpenAI-compatible

Both support streaming (SSE). Format translation is automatic and bidirectional.

Provider Routing

claude-* → Anthropic
gpt-* → OpenAI
gemini-* → Google
grok-* → xAI
org/model → Auto-routed
Override via X-Nyquest-Base-URL
// Reliability

How Optimization Reduces Hallucination Risks

Structured prompt optimization reduces the conditions that cause models to hallucinate.

Hallucinations increase when:

Prompts are ambiguous or instructions buried in verbosity
Constraints are diffuse — spread across paragraphs
Context window fills with redundant or contradictory history
Critical instructions evict silently at token limits

Nyquest addresses each vector:

Constraint canonicalization extracts and normalizes buried constraints
Conflict detection finds and resolves contradictory instructions
Redundancy pruning collapses repeated instructions into single form
Context optimization keeps important instructions in the window
// The Concept

Signal Processing for Language

Same principle as audio compression. Preserve the frequencies that carry meaning. Remove the noise that doesn't.

● Original Signal
● Compressed Signal

See it in action

Open the app and send a message. The optimization engine fires on every request automatically.

Open Nyquest All Products