The Nyquest App is the primary product. Around it: a public REST API for developers, a self-hosted engine for air-gapped deployments, and a browser extension. Pick what fits how you work.
// Primary Product
Nyquest App — app.nyquest.ai
LIVE
A full AI workspace. Talk to models from every major provider through one interface. Smart routing across tiers, model pinning, web grounding, persistent memory with vector recall, agent mode, artifacts (14 file types), audio + image generation, and ten distinct visual chassis — all in one place.
Free has access to BYOK plus whichever models are currently no-cost in the catalog. Pro unlocks every model — Claude Sonnet 5 · Fable 5 · Opus 4.8, GPT-5.6 Luna Pro · Luna · Terra Pro, and the rest. Circuit breakers fail over between tiers automatically; you always know which model handled which request.
Model Pinning
Lock any conversation to a specific model. Your pinned model handles every message in that thread — no routing surprises, no model switching mid-conversation.
BYOK
Bring your own API keys for OpenAI, Anthropic, Google, xAI. Keys are AES-256-GCM encrypted at rest and used for direct provider calls. Your rates, your billing.
Web Grounding
Live web search and full-page fetch injected into context before the model responds. No Tavily key required — built in. Responses cite the URLs they used.
Memory + Recall
Conversations persist. A separate memory system extracts facts across sessions, sanitizes PII, and recalls relevant ones via vector search at chat time. Confidence-scored.
Agent Mode
Multi-step planner that decomposes a request, runs tools (search, fetch, artifact creation, memory query), and streams steps over SSE. Per-run guardrails on cost, time, and steps.
// Create
Generate images and audio in chat.
Type a prompt, get an image. Ask for speech, get audio. Nyquest detects creative intent and routes to the right model automatically.
🎨
Image Generation
Describe what you want. Nyquest detects image intent and routes automatically. Images appear inline in your conversation and saved to your gallery.
🔊
Text-to-Speech
Generate natural-sounding speech from text. Multiple voices. Long-form inputs are auto-chunked. Audio files saved to your account and playable in the browser.
🖼️
Studio Interfaces
Dedicated Image Studio and Audio Studio for focused media work. Choose models, adjust settings, and manage generated files in one place.
// Artifacts
Upload files. They become context.
Drop a file into a conversation. Nyquest processes it automatically — extracts text, structures data, generates previews — and injects it into the model's context. 14 file types, 50+ extensions accepted. Per-user 100 MB media cap.
PDF
Text extracted automatically. Full document content injected as context.
Office Docs
Word, Excel, PowerPoint processed. Tables and structure preserved.
Code
Source files in dozens of languages. Indexed and made available verbatim to the model.
JSON / NDJSON
Parsed and structured. Schema detected. Key-value data available to models.
CSV / Logs
Tabular data parsed. Log files indexed. Plain text included directly.
Images
Passed to vision-capable models. Used for analysis, OCR, and visual reasoning.
Audio
Audio files transcribed and analyzed. Generated TTS clips saved to your account.
Plus archives, structured data formats, and more. Async processing for large files.
// Models + Routing
All major models. One interface.
332 models available through a searchable selector. Pin a specific model or let the dispatcher route through tiered defaults. Bring your own API keys for direct provider billing.
every request routed by complexity, domain & live price
🧭
Tiered Routing
T1 handles everyday work fast. T2 brings frontier-class models for Pro. T3 takes on deep reasoning and heavy workloads. Circuit breakers fail over between tiers automatically — every routing decision is visible and overrideable.
📌
Model Pinning
Lock a conversation to a specific model. GPT-5.6 Luna Pro, Claude Sonnet 5, Gemini 3.6 Flash, Grok 4.5 — whatever you want. The pinned model handles every message.
🔑
BYOK Supported
Bring your own API keys. Requests go directly to the provider under your account. Your rates, your billing. Keys encrypted with AES-256-GCM.
Nyquest searches the web and fetches actual page content. Results are injected directly into model context before the response is generated.
🌐
Search + Fetch
Runs a web search, then fetches full page content. Sandboxed, content-type gated, size-capped. Sources logged for provenance.
📰
Current Information
Ask about today's news, current prices, recent events. Responses grounded in live data, not stale training data.
🔓
No API Keys Required
Web grounding is built into the platform. No Tavily key. No SerpAPI subscription. Toggle on and go.
// Memory
Conversations don't reset.
Full conversation history stored. A separate memory system extracts facts, sanitizes PII, scores them by confidence, and recalls the relevant ones via vector search at chat time.
💬
Conversation Memory
Every conversation is stored. Auto-generated titles. Pick up where you left off. Soft-deleteable, searchable, organizable into projects.
🧠
Fact Recall
Embedding-indexed fact store. At chat time, your message is embedded and the top-K relevant facts are vector-searched and injected into context.
🔄
Confidence Scoring
Newer and more-corroborated facts rank higher. Stale facts decay. PII and credential-shaped numerics are scrubbed before storage.
// Agents
Multi-step agents, in the workspace.
Hand off a goal — not just a prompt. Agent mode plans, executes tools, and reports back. Steps stream live, every action is auditable, and per-run guardrails cap cost and time.
🧩
Planner + Runtime
The planner decomposes your goal into steps. The runtime executes each step, with the tool registry deciding what tools (search, fetch, artifact creation, memory query) the agent can call.
📡
Live SSE Stream
Watch each step happen in real time. The agent timeline shows tool calls, intermediate reasoning, and partial results as they arrive — no polling, no surprises.
🛡️
Guardrails
Every run has step, time, and cost caps. You can stop a run at any moment. Per-run state and per-step records are persisted — fully auditable.
Agent mode is in beta. Available on every plan.
// Analytics
See what the system is doing.
Every request is logged. Every routing decision is visible. You always know which model handled a request, how many tokens it used, and what it cost.
📊
Usage Tracking
Per-request cost, tokens in/out, model used, latency. Broken down by day, by model, by conversation.
🧭
Routing Visibility
See which tier was selected, which model actually served the request, and why. No black-box decisions.
💵
Billing Transparency
Full transaction history. Wallet balance, charges, top-ups. Every dollar accounted for.
// Infrastructure
Built as a real system, not a demo.
Purpose-built backend. Persistent storage. File processing pipelines. Background jobs. Daily backups. This is production infrastructure.
Rust/Axum Backend
Async, low-latency
Persistent Storage
Postgres, 36 tables
File Processing
14 types, 50+ ext.
Async Jobs
Per-artifact queue
Daily Backups
Automated
AES-256-GCM
Key encryption
SSE Streaming
Real-time tokens
332 Models
All major providers
// Chassis & Customization
12 chassis. Same workspace, different soul.
Same data, same flows, twelve radically different aesthetics. Each chassis is a complete visual identity — its own typography, palette, copy, and ambient effects. Switch any time. All chassis are included.
apsis
Map the system architecture
Charting the topology now — three layers, six modules, two upstream gateways…
apsis://prompt
apsis
cosmic terminal
Indigo and cyan glass. Stars in the background. The default for users who want depth and quiet light.
atelier
Sketch a colour study for late afternoon
Vermilion ground, warm umber shadows, a single cool note in the upper third…
brush in something…
atelier
painter's studio
Warm cream and vermilion. Italic accents. For long-form thinking and slow drafts.
atlas
Plan a route across the western states
North on the Pacific corridor, east through the Sierra passes, south along the Continental Divide…
Where to next?
atlas
mid-century guide
Cobalt with mustard yellow accents. Confident, navigational. For planning and structured work.
courtroom · CR-9714
Cite controlling precedent on the matter.
Pursuant to the holding in Re Westman, the relevant standard is…
Submit pleading…
courtroom
legal brief
Parchment and oxford blue. Serif throughout. For research, citations, and careful argument.
garden
When should I prune the climbing rose?
Late winter, before bud break — remove dead canes first, then shape the leader…
tend to a thought…
garden
botanical journal
Sage and lavender. Soft, organic, hand-bound feel. For reflection and personal notes.
matrix · tty1
$ explain entropy in compression
> entropy bounds the minimum bits needed to encode a signal…
$ _
matrix
hacker terminal
Green-on-black. Monospace everything. CRT scanlines. For code and command-line minds.
nyquest · ENGINE
Route this to the best model.
Compressed, dispatched, answered — teal on near-black…
ask anything…
nyquest
brand · default
Near-black surfaces, teal accent, Outfit + DM Mono. The Nyquest brand — and the chassis you start on.
paper · MORNING ED.
Headline this for the front page.
SHIPPING LANES SHIFT — analysts cite three converging forces…
file the next column…
paper
morning broadsheet
Warm off-white and terracotta. Serif headlines. For writing, editing, and journalistic drafts.
prism · 2036
Open a new array.
One running spectrum, everything else instrument-panel quiet…
start a stream…
prism
2036, not 2026
Rainbow LED filament on deep violet-black. A single running spectrum, used with restraint. The future chassis.
Front-month vs spot is 0.42% — within range. The carry argument holds if vol stays under 18…
desk > type order…
quant
trading desk
Dark navy and gold with a live ticker. Finnhub-backed market data. For numbers, analysis, and decisions.
ridgeline
Summarize this week's customer feedback.
Three themes: pricing clarity (5 mentions), onboarding speed (4), and integration depth (3)…
Ask anything…
ridgeline
premium SaaS
Clean white with teal accents. Familiar and crisp. For work that's going in front of clients.
studio · 128 BPM
Cut the second verse and add a bridge.
Bouncing the new arrangement — 32 bars, modulating up a half-step at the bridge…
▶ track 03 :: ready
studio
DAW / producer
Deep dark with magenta and lime. High-contrast, energetic. For creative work and rapid iteration.
Same data, different surface
Every chassis renders the same conversations, the same memory, the same artifacts. Switching is purely visual — no state migration, no settings to re-enter.
Per-chassis copy + ambience
Each chassis ships its own templates and ambient overlays — quant has the live ticker, matrix has scanlines, apsis has a starfield. Beyond palette, the workspace feels different.
WCAG-AA throughout
Every chassis is contrast-tuned for readability. Even matrix and studio meet AA on body text. No theme is for show only.
All 12 chassis included on every account. Switch any time from the chassis modal.
// BYOK — Bring Your Own Keys
Use your own API keys. Pay provider rates directly.
BYOK is a first-class feature. Plug in your API keys for any supported provider and your requests go directly to that provider under your billing. No token resale. No middleman markup. No surcharge.
🔐
AES-256-GCM at rest
Keys are encrypted the instant you paste them. Per-key nonce stored alongside ciphertext. Decrypted in memory only at dispatch. Never logged, never cached in plaintext, never sent anywhere except the provider you chose.
🔀
Direct provider passthrough
With BYOK, your request goes directly from Nyquest to the provider's API using your key. You pay the provider's published rate — or whatever volume discount you've negotiated. Nyquest adds zero token markup.
⚙️
Per-provider adapters
Each provider gets its own adapter handling auth, request shape, and streaming quirks. You can mix BYOK for one provider with hosted billing for others — the dispatcher routes accordingly.
// BYOK Flow
Your Prompt
│
├── Your API Key ──────── decrypted in memory · used once · never stored in plaintext
│
├── Provider API ──────── direct HTTPS call to OpenAI / Anthropic / Google / xAI
│ billed to YOUR account at YOUR rate
│
└── Response streams back ─ through Nyquest → to your workspace
Supported providers
OpenAI (GPT-5.6 Luna Pro · Luna · Terra Pro) · Anthropic (Claude Sonnet 5 · Fable 5 · Opus 4.8) · Google (Gemini 3.6 Flash · 3.5 Flash-Lite · 3.5 Flash) · xAI (Grok 4.5 · Build 0.1 · 4.3) · Any compatible endpoint via custom base URL.
Enterprise-friendly
Already have enterprise API agreements with volume pricing? BYOK preserves your negotiated rates. Committed-use discounts, reserved capacity, org-level billing — all stay intact.
Key management
Add, rotate, or remove keys any time in Settings → Providers. Each key is scoped to its provider. Mix and match BYOK and hosted billing per provider.
// Public API
Public REST API — api.nyquest.ai
LIVE
Same routing, same models, same wallet — accessible programmatically. Issue an API key from your account and integrate Nyquest directly into your apps, agents, or scripts.
OpenAI-compatible
POST /v1/chat/completions with streaming SSE. Drop-in replacement for any OpenAI-compatible client — just point base_url at https://api.nyquest.ai and pass your nq-v1-... key.
Stateless by design
API-key requests skip memory recall and conversation persistence. Each call is independent — no surprising prior-context bleed. Mirrors industry-standard developer-API behavior.
Same wallet, same routing
API requests draw from your wallet balance and route through the same dispatcher tiers as the app. Pin a model with model=... or let the router choose. BYOK works here too.
Manage keys in Settings → API Keys. Per-key usage and rate-limit tracking. Revoke any time.
// Self-Hosted
Compression Engine — Open Source, MIT/Apache-2.0
v3.2
The compression engine that powers the platform is open-source. Run it on your own infrastructure, behind your own firewall. Prompts never leave your network. Built for privacy-sensitive and air-gapped environments. v3.2 adds a published Docker image — pull from GHCR or build from source.
Identical to what runs behind app.nyquest.ai. Six-stage pipeline: normalize, OpenClaw agent mode, cache reorder, regex compression (532 rules across 19 categories with RegexSet prefilter and bounded LRU cache), local LLM semantic stage, auto-scale + forward. Sanitized for public release.
Docker or bare metal
Multi-stage Dockerfile (Rust builder + Debian-slim runtime, non-root user, tini PID-1) auto-published to ghcr.io/nyquest-ai/nyquest-engine on every release tag. Or build the single Rust binary from source with the one-shot installer. Runs on a Raspberry Pi (rules only) or with GPU for the semantic stage.
Open source
MIT/Apache-2.0 dual license. Nyquest-ai/nyquest-rust-fullstack-3.2.0. GitHub Actions CI · 47 / 47 tests passing · Docker auto-publish on tag. Privacy mode by default — no telemetry, no phone-home, prompts never logged.
Chrome extension that intercepts prompts on Claude, ChatGPT, Gemini, and Grok. Applies 281 compression rules locally before send — no proxy, no server, no account required.