SETUP

Install the local CLI, drop in the vLLM proxy, or connect via Claude Code.

Install CLI

The NeuroFS CLI is a local terminal client with guided onboarding, an ingest playground, plan/endpoint status, and dashboard visualization. Install with one command.

macOS / Linux

$curl -fsSL https://neurofs.com/install.sh | sh

Windows (PowerShell)

$iwr -useb https://neurofs.com/install.ps1 | iex

Then launch

$neurofs

Start in demo mode instantly, or paste your API key to unlock your plan.

CLI Commands

neurofsLaunch the interactive TUI

neurofs ingest --user u1 "prompt"Ingest and route a prompt

neurofs statusShow plan, endpoint, rate limits

neurofs loginAuthenticate with your API key

neurofs region listList all semantic regions

neurofs state --user u1View current activation state

neurofs config getShow daemon configuration

neurofs updateSelf-update to latest release

vLLM / LoRAX Proxy

OpenAI-compatible proxy that sits between your clients and a vLLM or LoRAX inference server. Every request is routed through NeuroFS, which selects the best LoRA adapter and injects it automatically — no client changes needed.

1. Install

$pip install neurofs-vllm-proxy

2. Configure

export NEUROFS_HOST=https://api.neurofs.com   # or your personal endpoint
export NEUROFS_API_KEY=nfs_...
export VLLM_HOST=http://localhost:8000
export VLLM_MODEL=llama3

3. Run

$neurofs-proxy

Listens on :9000. Point your clients at the proxy instead of vLLM directly.

What happens on each request

Proxy extracts the user prompt from the request
Calls NeuroFS to get the optimal adapter and routing plan
Injects the adapter into the forwarded request
Streams the response back to the client

Falls back to the original request if routing fails or times out (2s default).

Optional features

A/B testing

Shadow or canary modes via NEUROFS_MODE

MoE routing

Expert-group hints via EXPERT_ROUTING_ENABLED

LoRAX mode

Switch backend via BACKEND_MODE=lorax

Claude Code Integration

Add NeuroFS routing to Claude Code with a single hook. Before every tool invocation, the hook routes the current message through NeuroFS and injects the top regions and eligible tools as context — automatically.

1. Set your environment

export NEUROFS_HOST=https://api.neurofs.com
export NEUROFS_API_KEY=nfs_...

2. Add the hook to your Claude Code config

{
  "hooks": {
    "PreToolUse": [
      {
        "command": "python integrations/claude_code_hook.py",
        "timeout": 3000
      }
    ]
  }
}

The hook fails silently and never blocks the LLM call (2s timeout). It injects a YAML block with top regions, eligible tools, and token budget.

OpenClaw Setup

Authentication

All endpoints (except /api/v1/health) require an API key supplied via one of:

x-api-key: YOUR_API_KEY header
Authorization: Bearer YOUR_API_KEY header

The CLI stores your API key in OS-native secure storage (macOS Keychain, Linux Secret Service, Windows Credential Manager). The vLLM proxy reads it from NEUROFS_API_KEY.

Generate your API key from the Dashboard → API Keys section (requires sign-in). See pricing for per-plan rate limits.

Rate limits are enforced server-side per API key using a sliding 60-second window. Free tier: 20 rpm · Starter: 60 rpm · Pro: 200 rpm.