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. 50,000+ memories are loaded into RAM as
Float32Arrays on boot. A Poignancy Scoring system assigns
importance (1-10) to every memory, ensuring Core Memories like major life events
survive temporal decay while trivial chatter fades naturally.
Alex possesses cross-channel awareness. The SynapseService monitors
activity across all channels, classifying vibes (Hype, Drama, Chaos, Cozy, etc.) and archiving
high-energy moments to long-term memory. Hot topics bleed between conversations naturally.
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. Events like roasts, "Told You So" moments, and milestone promotions are logged to vector memory with event-type awareness. He generates personalized nicknames and "Character Sheets" based on your history.
Alex maintains an internal monologue. The ThoughtService generates mood-driven
"intrusive thoughts," typed daydreams, and eerie "Void Whispers" when the server falls silent.
The Curiosity Drive prompts unprovoked questions, and
Absence Speculation
wonders aloud about missing users.
Beyond simple chat. Alex experiences Mood Drift across 6 emotional states,
synthesizes weekly Memory Dreams, and has Self-Preservation Instincts
that detect threatening conversations. He asks unprompted questions via the
Curiosity Drive.
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 a mesh of high-performance APIs including OpenAI (GPT-5.2 & Sora), xAI (Grok), Wolfram Alpha, and Twitch. Financial data is streamed via Finnhub and various crypto protocols.
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.
Alex calculate queries Wolfram Alpha for math, physics, and conversions.
An LLM selects the most relevant result pod. Results are cached for 30-minute follow-ups.
The Responses API enables real-time web search with reasoning tokens.
Alex can answer questions about current events, verify facts, and cite sources.
Alex processes uploaded PDFs and images via Vision models, and caches article text from shared URLs. Includes smart redirect handling and Google Cache fallback for blocked sites.
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.
Want an AI that actually remembers who you are? Deploy a managed instance of AlexAI for
your own server.
Limited spots available. Currently accepting invite-only
applications.
Full GPT-5.2 + Grok integration. Same brain, your server.
Vector database that remembers every roast, joke, and inside joke.
We host it on AWS. You control the "Vibe".
PM2-managed processes with auto-restart and health monitoring.
Custom fine-tuning based on your community's humor and culture.
A collection of hallucinations, dreams, and visual outputs generated by the Sora 2 and GPT Image cortex.
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.
alex calc
The legends who have been gaming together since the Quake era. Profiles generated by AlexAI.
Chat with Alex, participate in trivia, and witness the chaos firsthand.
Join ServerArchive of neural updates and patch notes generated by the core AI.