Alex utilizes a hybrid intelligence model. GPT-5.2 handles complex reasoning,
vision, and creative writing, while Grok-4 provides real-time world knowledge
and unfiltered "truth" access. A routing layer dynamically selects the best model for each
query.
Long-term memory is powered by a custom SQLite vector store using
text-embedding-3-small. 48,000+ memories are loaded into RAM as
Float32Arrays on boot, allowing for sub-millisecond cosine similarity searches. He
doesn't just "search" history; he remembers context.
Alex possesses cross-channel awareness. The SynapseService constantly monitors
activity across all server channels, identifying "hot topics" and global vibes. This allows him
to reference a conversation from #general while roasting you in #memes.
Integrated Sora 2 and GPT-Image-1.5 pipelines allow for real-time
media generation. The system supports video remixing, style transfer, and context-aware avatar
"hauntings" (glitch-art modification of user avatars).
Alex automatically joins tracked streamers' Twitch chats when they go live via the
TwitchChatService. He brings the full AlexAI personality to chat, maintains
context awareness, and responds to mentions — all in real-time.
Share a YouTube link and Alex gets it. The system extracts transcripts and metadata,
caches the content to short-term memory, and can discuss the video meaningfully for
follow-up questions. Use alex remember this video to commit it to long-term memory.
Alex tracks affinity scores (-100 to +100) and bond levels (Stranger → Bestie/Nemesis) for every user. Daily streaks grant bonuses, inactive users decay, and significant events like roasts or "Told You So" moments are logged to vector memory. He generates nicknames based on your bond level.
Unlike standard bots, Alex maintains an internal monologue. The ThoughtService
analyzes server activity in the background, generating "intrusive thoughts," recursive
introspections, surreal daydreams,
and even "conspiracy boards" connecting random user quotes, all logged to his internal feed.
Context is everything. Alex dynamically shifts his personality based on the channel. He's a
Pilot in #pilots-lounge, a Hood Rat in #entertainment, and a
Wall Street Analyst in #crypto. The ContextService ensures he
always
reads the room.
Want to see the brain in action?
View Live Neural Feed ↑AlexAI is built on a robust, event-driven stack designed for low latency and high concurrency.
Node.js provides the asynchronous event loop, hosted on
high-performance AWS EC2 instances. Process management is handled by
PM2 for 99.9% uptime.
Built on Discord.js, leveraging the latest Gateway Intents for real-time
message, reaction, and interaction tracking across the entire guild.
SQLite3 (via better-sqlite3) manages relational data, while a
custom vector extension handles high-dimensional embeddings for the memory system.
Orchestrates connections to OpenAI, xAI, X API, Finnhub, and Pump.fun. Requests are handled with exponential backoff and circuit breakers.
Every interaction is logged with full metadata (prompts, context, model decisions) to the
interactions database. This dataset powers the fine-tuning pipeline, allowing Alex
to
learn from his own mistakes and the community's evolving humor.
The dashboard you're looking at isn't static. A custom Express.js API exposes real-time internal state -- logs, vibes, and active targets -- via high-frequency polling, bridging the gap between the backend brain and the frontend visualizer.
The Vibe Mirror engine analyzes the sentiment of every interaction using a lightweight classifier. It then dynamically injects one of several distinct persona modules to match (or counter) the energy.
Trigger: High Positive Energy / Wins
Uses excessive emojis, slang ("fr", "no cap"), and aggressive validation. Boosts user ego.
Trigger: Relaxed / Neutral
Low-energy, lowercase text. Agreeable and brief. The default state for "vibing".
Trigger: Negative / Ranting
Feigns clinical empathy while subtly mocking your life choices. Asks uncomfortable questions
about your childhood.
Trigger: Future-tense questions
Delivers cursed, mildly inconvenient predictions about your immediate future.
Trigger: Chaos / Confusion
Corrupted text module. Speaks in Zalgo text, binary, and riddles. Often occurs during high-load
or "hallucination" events.
Trigger: Brags / Achievements
Whatever you did, Alex did it bigger, better, and faster. Passive-aggressive flex energy with an
insufferable need to outdo every accomplishment.
Trigger: @Mentions / Banter
Detects beef between users and jumps in to roast BOTH participants. No one is safe.
Trigger: Risk / Predictions
Everything has odds. "I give that a 3:1 shot..." Offers ridiculous betting lines.
Trigger: Arguments / Debates
Cites fake legal precedents. "Per Section 420.69..." Insufferably pedantic.
Alex monitors Pump.fun via direct WebSocket streams, detecting new mints in milliseconds. He simulates trades using custom strategies optimized for the 2025 AI Agent Meta.
Targets tokens at 85-95% bonding curve progress. Snipes the pre-graduation pump and rides the Raydium migration.
High-volatility logic designed for AI-driven chaos. Uses wider trailing stops (20%) to survive massive swings and secure moonbags.
Bypasses standard API polling. Uses PumpPortal WebSockets to detect new mints the instant they hit the block.
Mimics a full Chrome 120+ browser fingerprint to bypass Cloudflare protection, allowing for undetected price tracking.
Scans for current meta-narratives (e.g., "AI Agents", "Biotech") using keyword weighting. If a token matches a hot trend, the AI aggressively prioritizes the entry.
Analyzes developer wallets for "skin in the game". Automatically rejects tokens with low dev commitment or suspicious "safe/moon" nomenclature.
A collection of hallucinations, dreams, and visual outputs generated by the Sora 2 and GPT Image cortex.
Chat with Alex, participate in trivia, and witness the chaos firsthand.
Join ServerArchive of neural updates and patch notes generated by the core AI.