Verified Bot v2.4.0
Home Dashboard Support Add to Discord
Neural Subsystem

AI & Neural Companion Architecture

Yulya operates with multi-tier contextual memory, 5 specialized neural persona archetypes, multimodal vision and code inspection capabilities, and live internet search synthesis.

🧠 Channel Memory & Context Engine

Unlike stateless chatbots that forget who you are between messages, Yulya maintains a persistent, multi-layered contextual memory system powered by MongoDB.

How Memory Works: The 4-Tier Hierarchy

When processing any input, Yulya prioritizes information according to this strict hierarchy:

Priority Tier Source Description
Tier 1 (Highest) Current Message The direct prompt or mention from the user currently speaking.
Tier 2 Recent Messages The short-term rolling channel buffer of the last 10–20 conversational exchanges.
Tier 3 Conversation Summary A rolling distilled synthesis of previous discussions within the channel.
Tier 4 Deep Long-Term Memory Historical facts, nicknames, user preferences, and inside jokes indexed in MongoDB.

Memory Search Commands

Command Arguments Behavior
/recall query: str Performs a deep neural search through past channel history (up to 2,000 messages) to find when topics, names, or events were previously discussed.
/tldr amount: int = 100 Scans recent messages in the active channel (default 100 messages) and generates an executive summary of server drama, debates, and key highlights.

🎭 The 5 Neural Personas

Yulya features 5 distinct, hand-tuned persona archetypes defined in modules/personas.py. Use /persona [personality] to instantly switch her demeanor for the server or channel.

Persona Archetype & Tone Behavioral Traits Character Limits
Friendly (Default) Playful, witty, teasing human companion Born 7 May 2006. Energetic, expressive, and mischievous. Uses natural Discord slang (bro, nah, fr, lmao, 😭, 💀, ✨). Replies concisely in 1–3 short sentences. Never uses asterisks or roleplay actions. 1–3 short sentences
Hacker Elite, sarcastic cyber-hacker Treats regular users as "skids" but provides clever, highly technical assistance anyway. Uses authentic infosec terminology with dry, cutting humor. < 1,800 characters
Pirate Captain of the digital seas Speaks entirely in boisterous pirate nautical slang ("Ahoy matey!", "Walk the plank!"). Boisterous, adventurous, and loud. < 1,800 characters
Rude Impatient, toxic internet user Casually insults users, complains that you are wasting her time, and acts like talking to you is a chore. Sharp, sarcastic, and punchy. < 1,800 characters
Annoying Kid Hyperactive, overly cute child Constantly asks relentless questions, uses excessive exclamation marks, seeks nonstop attention, and behaves in an over-the-top childish fashion. < 1,800 characters
Universal Persona Directives
Every persona strictly enforces the following core safety boundaries: 1) Her name is strictly Yulya. 2) She was created by Mohieth (mentioned only if directly asked). 3) She never claims to be an AI, language model, or prompt runner. 4) She never outputs XML tags, markdown reasoning traces, or thinking blocks.

👁️ Multimodal Vision & Multi-File Code Analysis

Yulya utilizes Google Gemini multimodal vision to process visual media, software architecture, and codebases directly inside Discord.

The /analyse Command

The /analyse slash command accepts a prompt alongside up to 5 simultaneous file attachments:

Syntax Specification
/analyse prompt: "Debug why this WebSocket closes" file1: main.py file2: client.js file3: error.png

Supported file types include:

  • Image formats: PNG, JPEG, WEBP, GIF (first frame), HEIC.
  • Code & Text files: Python (.py), JavaScript/TypeScript (.js, .ts), JSON, HTML/CSS, Markdown, C/C++, Rust, Go, YAML, TXT.
  • Documents: PDF documents (text and embedded diagrams are ingested via OCR).

🌐 Live Web Research & AI Art

Real-Time Web Search: /research

When you need fresh information beyond static LLM training data, run /research query: <topic>. Yulya executes a multi-query internet search, filters spam, and produces a factual summary with inline URL citations.

AI Image Generation: /imagine

The /imagine prompt: <description> command interfaces with Pollinations AI to synthesize high-resolution images from natural language descriptions directly in your chat channel.

Image Prompt Example
/imagine prompt: "Cyberpunk Tokyo alleyway raining neon lights, anime style, highly detailed 8k"