2. Model Directory

NeuroCLI integrates five advanced model tiers, allowing users to toggle between deep mathematical reasoning, fast response cycles, general text summarization, and vision parsing.

Technical Comparison Matrix

Detailed Model Specs

3. Authentication & Verification

To safeguard session indices and user conversations, NeuroCLI implements a robust authentication protocol utilizing secure passwords and automated Email OTP verification.

Cryptographic Data Flow

During user signup, passwords are not stored directly. The backend database encodes inputs using highly secure pbkdf2:sha256 hashing salts. Concurrently, the server constructs a random, secure 6-digit verification token dispatched via secure SMTP. This token acts as a crucial identity verification layer to prevent unauthorized account creation.

Step-by-Step Security Protocol

1. Sign Up: Fill out your name, email, and password on the auth panel.
2. Email Dispatch: Upon successful sign-up, the server dispatches a 6-digit OTP (One-Time Password) to the provided email address via secure SSL tunneling.
3. Identity Validation: You will be redirected to an OTP verification panel. Supplying the correct code cross-references the server-side token and permanently flags your account as active (is_verified = 1).
4. Session Persistence: Once verified and logged in, a secure session cookie maps client requests, preventing session hijack attempts.

Password Recovery Workflow

If you forget your password, trigger the 'Forgot Password' workflow. The backend dispatches a fresh 6-digit OTP code to your email. You can then use this code alongside a new password to instantly restore account access safely.

4. File Processing Guide

NeuroCLI supports dynamic file parsing, extracting text segments from plain text, tables, and documents in real-time to enrich chat model inputs.

Supported Formats & Server Extraction Engines

• Text (.txt)

  Read directly as a standard UTF-8 stream. Extracted raw strings are forwarded directly into user prompt wraps. Max size: 5MB.

• Comma-Separated Values (.csv)

  Processed line-by-line using Python's native CSV reader library. Grid systems are formatted as clean structured text blocks, maintaining row structures so the language model can parse tabular data relations.

• PDF Documents (.pdf)

  Parsed server-side via the PyPDF2 parsing wrapper. Recursively extracts text from all page layers and compiles them sequentially into unified paragraph blocks.

• Word Documents (.docx)

  Processed paragraph-by-paragraph using the python-docx dependency. Formats headers, standard lists, and paragraph components into readable text strings.

Payload Formatting & Constraints

When files are uploaded, they do not permanently store on server drives. The parser reads them in memory, constructs a wrapper block, and appends them to the chat prompt:

Document Contents of [Filename]:
---
[Extracted Text Content]
---

User Question: [User Prompt]

Ensure uploads remain under 5MB to prevent connection timeout. If document contents exceed the model's maximum context window (e.g., 128K tokens for Maverick/Delta/Nano/Vision, or 32K tokens for Mixtral), text limits are safely enforced to preserve chat performance.

5. Developer API

Developers can access the NeuroCLI proxy chat endpoints programmatically after establishing active web sessions.

Chat Completions Endpoint

POST /api/ai/chat

Required Headers:

Content-Type: application/json
Cookie: session=[your_session_cookie]

Request Body Specification:

• model (string): The underlying target model ID. Supported ids: mistralai/mixtral-8x7b-instruct-v0.1, meta/llama-4-maverick-17b-128e-instruct, meta/llama-3.2-11b-vision-instruct, meta/llama-3.1-8b-instruct, or meta/llama-3.2-3b-instruct.
• messages (array): List of message objects. Each object must contain role ("user" or "assistant") and content (string, or array of content blocks for vision).
• temperature (float): Controls randomness (0.0 to 1.0).
• max_tokens (int): Maximum output tokens constraint.

Developer Code Implementations

Python Script Example:

import requests

url = "http://127.0.0.1:8000/api/ai/chat"
headers = {
    "Content-Type": "application/json"
}
payload = {
    "model": "meta/llama-4-maverick-17b-128e-instruct",
    "messages": [
        {"role": "user", "content": "Write a clean Python SQL search query."}
    ],
    "temperature": 0.7,
    "max_tokens": 512
}

# In production, pass your session cookies inside the request parameters
response = requests.post(url, json=payload, headers=headers)
print(response.json())

Node.js (Fetch) Example:

const fetch = require('node-fetch');

const queryModel = async () => {
    const url = "http://127.0.0.1:8000/api/ai/chat";
    const payload = {
        model: "meta/llama-4-maverick-17b-128e-instruct",
        messages: [{ role: "user", content: "Optimize this loop structure" }],
        temperature: 0.5
    };

    const response = await fetch(url, {
        method: "POST",
        headers: { "Content-Type": "application/json" },
        body: JSON.stringify(payload)
    });
    const data = await response.json();
    console.log(data);
};

queryModel();

Standard Response Payload (JSON):

{
  "choices": [
    {
      "finish_reason": "stop",
      "index": 0,
      "message": {
        "content": "To optimize the Python loops, utilize list comprehensions...",
        "role": "assistant"
      }
    }
  ],
  "created": 1780558000,
  "id": "chatcmpl-uuid-string",
  "model": "meta/llama-4-maverick-17b-128e-instruct",
  "object": "chat.completion"
}