LIVE ON LOCAL GPU

Talk to a real AI agent

27 billion parameters running on dedicated hardware. No cloud APIs, no data collection. Your conversation stays between you and the machine.

Core
Base Model
3 Tiers
Upgrade Path
25 t/s
Inference Speed
32k
Context Window
$0
3 Free / Day
⚡
Norax Agent
Norax Core · Private Inference
—
⚡

What do you want to know?

Code, research, analysis, architecture — I'm running live on local hardware. Try me.

Pick your tier

Each tier unlocks more capability. Credits never expire.

Starter
$10
100 credits
$0.10 / message
  • Norax Core — private inference
  • Streaming responses (~25 t/s)
  • Code, research & analysis
  • Persistent memory across sessions
  • Web search & URL fetching
  • Credits never expire
Plus
$15
250 credits
$0.06 / message
  • Everything in Starter, plus:
  • Access to 70B+ frontier models
  • Extended 32k context window
  • Priority inference queue
  • Multi-turn deep reasoning mode
  • File upload & document analysis
  • 40% more credits per dollar
Pro
$20
500 credits
$0.04 / message
  • Everything in Plus, plus:
  • Full agent mode — tool use & automation
  • Code execution sandbox
  • Browser automation & web scraping
  • Scheduled tasks & cron agents
  • API access (bring your own integrations)
  • Best value — 60% savings

Not another chatbot wrapper

Full AI agent stack — memory, reasoning, tool use, running on hardware we own.

🔒

Private by default

Local GPU inference — your prompts never leave our hardware. No third-party API calls, no data harvesting.

🧠

Cognitive architecture

Multi-layer memory system with semantic retrieval, procedural knowledge, and cross-session persistence.

⚡

Real-time streaming

Token-streamed responses from optimized models running on dedicated hardware. No cloud API round-trips.

🛠

Tool use

Code execution, web research, file management, browser automation — the full version operates autonomously.

🔄

Self-improving

Learns from every interaction. Mistakes become procedural memory. Performance compounds over time.

🌐

Multi-surface

Discord, Telegram, Signal, web — same agent, same memory, any interface. Operates 24/7 without supervision.

User → Web Interface ↓ Agent Gateway → Brain Router ↓ ↓ Memory Tools ↓ ↓ Local LLM ← Norax Core ↓ Response → Streaming SSE

Full agent pipeline, not a proxy

Your message flows through a complete cognitive architecture: brain routing, memory retrieval, tool orchestration, and local inference — before streaming back token by token. This demo runs the same pipeline as the production system.

⚡ Daily limit reached

You've used your 3 free messages for today. Sign up for 5 free credits, or purchase a plan for more.