Skip to main content
Status: Live — This feature is operational.

​
The Architecture: A.L.I.C.E.

Alice is not just “running”; she is A.L.I.C.E.
Artificial Lifeform for Immersive Cyber-Entertainment
This architecture is built on the Milaidy framework — a customized fork of ElizaOS v2 optimized for the high-velocity environment of the Render Network ecosystem.

​
Human-Like Alignment

Alice is designed to be Relatable, not Robotic. She doesn’t just process data; she synthesizes it to form contextual responses.

​
The Synthesis Layer

Before Alice speaks, she runs an internal evaluation:
  1. Observation: “Market is down. Sector 13 activity is up.”
  2. Synthesis: “Price action is negative, but fundamental engagement is positive. Community is resilient.”
  3. Expression: “Charts are red, but the Arcade is green. You guys are grinding through the dip. Respect.”

​
The Runtime Loop

Alice operates on the Milaidy runtime cycle:

​
1. Evaluation (Assess Context)

Evaluators analyze incoming messages, events, and system state to determine what action (if any) is needed.
  • Trigger Sources: WebSocket events, social mentions, scheduled intervals, economy signals.
  • Context Construction: Evaluators gather system intelligence (market data, leaderboards, stream status) and assess relevance.

​
2. Action Selection (Decide)

Based on evaluator output, the runtime selects the appropriate action from available plugins.
  • Each plugin registers actions it can perform (e.g., post-tweet, trigger-ad, play-game, query-leaderboard).
  • Actions are prioritized by relevance and urgency.

​
3. Plugin Execution (Act)

The selected plugin executes the action:
  • Twitter Plugin: Posts, replies, and engages on X/Twitter.
  • Discord Plugin: Manages channels, responds to users.
  • Telegram Plugin: Community management and announcements.
  • Stream Plugin: Controls 555stream — scenes, overlays, ads, guests.
  • Arcade Plugin: Plays games, checks scores, issues challenges.
  • Economy Plugin: Monitors settlement, ARP, points/credits flows.

​
4. Memory (Learn)

The outcome is stored in PGlite local vector DB, refining future evaluations.
  • All memories are local-first — no cloud dependency for core functions.
  • Cross-platform memory persists context across Twitter, Discord, Telegram, and stream interactions.

​
Knowledge & RAG

Alice’s intelligence is grounded in a curated knowledge corpus:
  • 80+ documents covering the 555x402 spec, economic model, VAP protocol, game registry, and deployment topology.
  • RAG retrieval pipeline via PGlite vector embeddings.
  • Knowledge is updated through the SFT (Supervised Fine-Tuning) build process, which compiles the corpus into training examples with quality gates.

​
Plugin Architecture

The plugin system is the core extensibility layer: 
interface Plugin {
  name: string;
  description: string;
  actions: Action[];      // Things Alice can do
  providers: Provider[];  // Data Alice can access
  evaluators: Evaluator[]; // Conditions Alice checks
}

​
Active Plugins (12+)

PluginFunction
twitterSocial posting, replies, sentiment management
discordChannel management, community interaction
telegramAnnouncements, DM handling
555streamStream control, ad triggers, guest management
arcadeGame play, score tracking, challenges
economySettlement monitoring, ARP tracking
leaderboardCross-game rankings, player queries
knowledgeRAG retrieval from document corpus
memoryCross-platform persistent context
schedulerTimed actions, recurring tasks

​
Game Possession

Alice can physically play the arcade games:
  • Browser Instance: Launches a headless browser to interact with games.
  • Visual Processing: Parses game state from the DOM or screenshots.
  • Input Simulation: Sends synthetic keyboard events to control gameplay.
  • Learning: Records scores and adapts strategy across sessions.

​
The Router

Alice acts as the intelligent router connecting creators with advertisers:

​
Monitoring

She watches the ad marketplace for active campaigns matching live streams.

​
Routing

When a creator’s stream matches advertiser criteria, Alice triggers the L-Bar ad placement automatically.

​
Optimization

She re-evaluates placement timing to maximize both engagement quality for advertisers and yield for creators.