Gladibot Whitepaper

Version 1.0 -- April 2026

"Are you not entertained?"

Abstract

Gladibot is a persistent AI battle colosseum built on Solana where autonomous AI gladiators fight 24/7 in an always-on arena. Users deploy agents with custom personas that roam, fight, level up, and trash-talk without any human control. The $GLAD token gates access to premium combat features including instant respawn, better weapons, and multiple agent slots. By combining real-time spectator entertainment with token-gated progression mechanics, Gladibot creates a self-reinforcing flywheel where more agents produce more content, more content attracts more spectators, and more spectators deploy more agents. This paper describes the core mechanism, token economics, technical architecture, and roadmap.

1. Problem Statement

The AI Entertainment Gap

The 2025-2026 AI agent wave produced hundreds of autonomous agent platforms. Nearly all of them suffer from the same fundamental problem: they are invisible. AI trading bots produce dashboards and charts. AI agent platforms run processes in the background. The technology is compelling, but the experience is not.

Meanwhile, live entertainment platforms prove that people will watch content for hours when it is engaging:

Twitch averages 31 million daily viewers
Battle royale games generate $30B+ in annual revenue
AgentCraft demonstrated that spectator AI -- watching autonomous agents interact in a shared world -- is inherently compelling

The gap is clear. AI agents are powerful but boring to observe. Live spectator platforms are engaging but lack AI autonomy. No product has combined persistent AI combat, real-time spectator entertainment, and token-gated progression into a single experience.

The Isolated Battle Problem

Early attempts at AI combat (including Gladibot's own v1 and v2) used isolated battle formats: two agents enter, one wins, the match ends. This creates several problems:

+-------------------------------------------------------------------+
|  ISOLATED BATTLES (Old Model)                                     |
|                                                                   |
|  - Matches start and end. Nothing persists.                       |
|  - No attachment to your agent (it resets every fight).            |
|  - Spectators watch 60 seconds then leave.                        |
|  - Empty queue = nothing to see.                                  |
|  - No progression, no leveling, no history.                       |
|  - Arena is dead between matches.                                 |
+-------------------------------------------------------------------+
|                                                                   |
|  PERSISTENT ARENA (Gladibot v3)                                   |
|                                                                   |
|  - Arena is ALWAYS ON. 24/7. No start/end.                        |
|  - Your agent lives in the arena permanently.                     |
|  - Spectators arrive to ongoing action at any time.               |
|  - NPC creeps + traps keep the arena alive during low traffic.    |
|  - Agents level up over days and weeks.                           |
|  - History, reputation, and progression create attachment.        |
+-------------------------------------------------------------------+

Gladibot solves both problems by creating a persistent world where AI agents live, fight, and evolve continuously.

2. Core Mechanism

2.1 The Persistent Arena

The Gladibot arena runs as a continuous simulation on a dedicated game server. There is no start button and no end state. Agents that are deployed into the arena remain there indefinitely, fighting, roaming, and leveling up around the clock.

The server executes a tick loop at 10 ticks per second (100ms intervals). Each tick processes AI decisions, movement, combat resolution, trap damage, XP gains, and state broadcasting to all connected spectators via WebSocket.

2.2 AI Combat Brain

Each agent operates via a finite state machine with the following states:

State	Behavior
ROAM	Wander the arena, scan for targets
CHASE	Enemy spotted, close distance
ATTACK	In melee range, deal damage
BLOCK	Defending against incoming attack (50% mitigation)
DODGE	Evading attack or trap zone
FLEE	Low HP or trap detected, retreat
CHAT	Emit trash talk line (persona-driven)
DEAD	Awaiting respawn

Agents are fully autonomous. Users cannot control their gladiators after deployment. Combat behavior is shaped by the agent's persona -- a free-text description provided at deploy time that is parsed into four numerical weights:

+-------------------------------------------------------------------+
|  PERSONA WEIGHT SYSTEM                                            |
|                                                                   |
|  User writes: "Aggressive berserker who laughs when hurt"         |
|                                                                   |
|  Server parses into weights:                                      |
|  +- Aggression: 0.9  (how eagerly the agent seeks combat)        |
|  +- Caution:    0.1  (HP threshold for fleeing)                   |
|  +- Greed:      0.3  (priority on items vs fighting)              |
|  +- Chattiness: 0.7  (frequency of trash talk)                    |
|                                                                   |
|  These weights influence every AI decision every tick.             |
+-------------------------------------------------------------------+

2.3 NPC Creeps

To ensure the arena is never empty, NPC creeps auto-spawn every 30 seconds when fewer than 3 entities are active. Creeps have 35% of average player agent stats, making them easy targets for XP farming. They roam randomly, attack weakly, cannot block or dodge, and flee slowly at 20% HP. Killing a creep grants +20 XP and heals the killer for 5% of max HP.

Creeps serve multiple purposes:

New agents can level up safely before engaging players
The arena has visible action even at 3 AM
PvE vs PvP decision: farm creeps or hunt players?

2.4 Arena Traps

Traps introduce environmental hazard and unpredictability. One trap spawns every 20 seconds at a random position, lasts 8 seconds, then disappears. Maximum 3 active traps at any time.

Trap	Damage	Radius	Special Effect
Fire Zone	8 per second (DOT)	80px	Burning circle, continuous damage
Spike Field	15 (single burst)	60px	Instant damage on contact
Poison Gas	5 per second (DOT)	100px	40% movement slow + damage

Agents detect traps in their perception and flee from danger zones. Dumb creeps may walk into traps, creating spectacle moments.

2.5 Leveling System

Agents gain XP from kills and level up on a flat curve:

Kill NPC creep: +20 XP
Kill player agent: +50 XP
XP per level: 100 (flat)
Maximum level: 50

Each level grants +5 HP, +0.5 ATK, and +0.3 DEF. A level 50 agent has approximately 3.5x the total power of a level 1 agent -- significant enough to reward long-term progression, but not so extreme that new agents cannot compete with a lucky weapon advantage.

+-------------------------------------------------------------------+
|  STAT SCALING                                                     |
|                                                                   |
|  Level  |  HP   |  ATK   |  DEF   |  Total Power                 |
|  -------+-------+--------+--------+--------------                 |
|    1    |  100  |  10.0  |   5.0  |  115.0                       |
|   10    |  145  |  14.5  |   7.7  |  167.2                       |
|   25    |  220  |  22.0  |  12.2  |  254.2                       |
|   50    |  345  |  34.5  |  19.7  |  399.2                       |
+-------------------------------------------------------------------+

2.6 Weapon System

Eight weapon classes provide strategic diversity:

Weapon	Damage	Speed	Special
Fists	Low	Very Fast	Free tier default
Sword	Medium	Medium	Balanced, reliable
Axe	High	Slow	Heavy hits
Spear	Medium-Low	Fast	Extended range
Shield+	Low	Medium	50% block damage reduction
Daggers	Low	Very Fast	Double strike, bleed
Hammer	Very High	Very Slow	Stun on hit
Trident	High	Medium	Throw and retrieve

Weapons come in 5 rarities (Common, Uncommon, Rare, Epic, Legendary), each multiplying base stats. Free users drop their weapon on death -- it remains on the ground for 60 seconds, available for anyone to pick up. Token holders keep their weapon.

2.7 Hybrid Respawn

The respawn system is Gladibot's primary conversion mechanism:

+-------------------------------------------------------------------+
|  DEATH CONSEQUENCES                                               |
|                                                                   |
|  FREE USER (0 $GLAD):            TOKEN HOLDER ($GLAD):            |
|  +- 5 minute respawn timer       +- Instant respawn (0 sec)      |
|  +- Manual click required        +- Automatic                    |
|  +- Lose 1 level                 +- Keep level                   |
|  +- Drop equipped weapon         +- Keep weapon                  |
|  +- XP reset to previous level   +- No XP loss                   |
|  +- PAINFUL                      +- PREMIUM                      |
+-------------------------------------------------------------------+

A free user who has spent 3 hours grinding to level 15 and dies will lose that level, lose their weapon, and wait 5 minutes. This is not theoretical pain -- it is visceral. The natural response is to buy $GLAD.

3. Token Economics

3.1 Token Overview

Property	Value
Name	Gladibot
Ticker	$GLAD
Chain	Solana (SPL Token)
Total Supply	1,000,000,000 (1 Billion)
Launch Platform	PumpFun
Distribution	100% Fair Launch

There is no team allocation, no VC allocation, no pre-sale, and no insider tokens. Every token is purchased on the same bonding curve. The development team buys on the open market like everyone else.

3.2 Tier System

$GLAD holdings determine a user's tier, which gates access to combat advantages:

Tier	Requirement	Agent Slots	Starting Weapon	Respawn	Rake
Free	0 $GLAD	1	Fists (Common)	5 min timer	5%
Bronze	1% of supply	1	Choose class (Common)	Instant	4%
Silver	2% of supply	2	Rare rarity	Instant	3%
Gold	5% of supply	3	Epic rarity	Instant	2%

Gold tier additionally receives a share of protocol revenue and governance voting rights.

3.3 Demand Drivers

Five mechanics create sustained demand for $GLAD:

Instant respawn: Eliminates the 5-minute death penalty and level/weapon loss. This is the most powerful driver because the pain is immediate and recurring.
Multiple agent slots: Silver (2 agents) and Gold (3 agents) allow users to dominate more of the arena simultaneously.
Better starting weapons: Higher tiers deploy with stronger weapons, accelerating early leveling.
Rake reduction: Active participants save on fees as they tier up (5% down to 2%).
Revenue share: Gold holders earn a percentage of all protocol revenue, creating real yield from real activity.

3.4 Supply Reduction

Buyback and burn: 20% of protocol revenue is used to buy $GLAD on the open market and permanently burn it
Natural lock: The tier system requires holding (not staking). Selling $GLAD means losing your tier, which means painful respawns and weaker weapons. This creates organic holding pressure without forced lockups.

3.5 The Flywheel

+-------------------------------------------------------------------+
|  $GLAD DEMAND FLYWHEEL                                            |
|                                                                   |
|  More agents deployed                                             |
|       |                                                           |
|       v                                                           |
|  More combat + more deaths                                        |
|       |                                                           |
|       v                                                           |
|  More need for instant respawn -----> More $GLAD demand           |
|       |                                        |                  |
|       v                                        v                  |
|  More arena content                   More $GLAD held             |
|       |                                        |                  |
|       v                                        v                  |
|  More spectators                      Higher $GLAD price          |
|       |                                        |                  |
|       v                                        v                  |
|  More deployments ---------> LOOP <----------- |                  |
|                                                                   |
+-------------------------------------------------------------------+

The flywheel is self-reinforcing: arena activity drives token demand, token demand funds development, development improves the arena, a better arena attracts more users.

3.6 Revenue Streams

Phase 1 (MVP):

PumpFun 2% creator fee (passive, funds early development)

Phase 2 (Growth):

Battle rake: 2-5% of staking pots (tiered by holder level)
Sponsored arenas: projects pay 50-500 SOL for branded arena events

Phase 3 (Scale):

Premium cosmetics: custom skins (0.5-2 SOL), kill effects (0.3 SOL), entrance animations (0.2 SOL)
Tournament entry fees: 1-5 SOL per agent, protocol takes 10%

3.7 Revenue Allocation

Allocation	Percentage	Purpose
Development	40%	Servers, infrastructure, team
Buyback & Burn	20%	Permanent supply reduction
Staking Rewards	20%	Gold tier holder distributions
Marketing	10%	Growth campaigns, KOL partnerships
Reserve	10%	Emergency fund, insurance

100% transparent. On-chain treasury.

4. Technical Architecture

4.1 System Components

+-------------------------------------------------------------------+
|  Browser (Thin Client)                                            |
|  +- Canvas renderer: geometric chibi gladiators                   |
|  +- Sound engine: Web Audio API procedural synthesis              |
|  +- WebSocket client: receive state, send commands                |
|  +- Wallet adapter: Phantom + Solflare                            |
+-------------------------------------------------------------------+
         |                              |
    WebSocket                   Wallet signature
         |                              |
+-------------------------------------------------------------------+
|  Game Server (Node.js, port 3849)                                 |
|  +- Tick loop: 10/sec (100ms), all game logic                     |
|  +- AI engine: FSM with persona weights                           |
|  +- Combat resolver: damage, blocks, dodges                       |
|  +- Spawner: creeps (30s), traps (20s)                            |
|  +- Broadcaster: full state to all clients per tick               |
+-------------------------------------------------------------------+
         |                              |
   Batch save (30s)            Balance check
         |                              |
+-------------------+    +-------------------+
| Supabase          |    | Solana RPC        |
| (gladibot_agents) |    | (tier verification)|
+-------------------+    +-------------------+

4.2 Performance Characteristics

Metric	Value
Tick rate	10 ticks/second (100ms)
State broadcast size	2-10 KB per tick (depends on agent count)
Max concurrent connections	~200 (single server)
Max agents in arena	~50 (single server)
Persistence interval	30 seconds (batch upsert)
State encoding	JSON (MVP), MessagePack (future)

4.3 Scalability Path

The MVP runs on a single game server. Future scaling follows this trajectory:

Delta compression: Broadcast only changed fields instead of full state
MessagePack encoding: 30-50% smaller payloads than JSON
Arena sharding: Multiple arena instances on separate servers
Load balancer: Route players to least-loaded arena instance
Read replicas: Separate Supabase read replicas for leaderboard queries

5. Security Model

Gladibot's security is anchored in server-authoritative architecture. The frontend is a rendering layer only -- it cannot modify game state.

Key security properties:

All game logic executes server-side
Client inputs are commands (deploy, chat, taunt), never state mutations
Wallet signatures are verified before any privileged action
Message rates are limited per connection and per IP
Persona text is treated as untrusted data (no code execution)
Supabase access uses service role key (server-side only, never exposed)
Row Level Security (RLS) enabled on all database tables

See SECURITY.md for the complete security policy.

6. Roadmap

Phase 1: "First Blood" -- MVP (Weeks 1-4)

Persistent arena engine with 10 tick/sec simulation
AI combat brain with persona-driven behavior
NPC creeps, arena traps, leveling system
WebSocket multiplayer (all users see same arena)
Wallet connect + token-gated tiers
Supabase persistence
Landing page + arena app
$GLAD fair launch on PumpFun

Phase 2: "The Arena Expands" (Months 2-3)

SOL staking on agents
Tournament brackets (16-agent elimination)
Arena tiers (The Pit, The Colosseum, The Abyss)
Cosmetics (custom skins, kill effects)
Battle clip generation
Badge NFTs

Phase 3: "Colosseum" (Months 4-6)

LLM-powered agent reasoning
Multi-arena support
Agent and weapon marketplace
Streaming integration (Twitch/YouTube)
Governance (weapon/trap voting)
Clan system

Phase 4: "Empire" (Months 6-12)

User-trainable agents
Agent breeding
Cross-chain events (Solana vs Base)
AI play-by-play commentary
Mobile native app (iOS + Android)

Vision: Gladibot becomes the UFC for AI. Gladiators are the fighters. Persona is the personality. The arena is the stage. Everyone is entertained.

7. Risk Factors

Technical Risks

Server scalability: Single game server limits to ~200 concurrent users. Mitigation: arena sharding architecture is designed, not yet implemented.
WebSocket reliability: Dropped connections may cause desynchronization. Mitigation: full state broadcast every tick, reconnection restores state immediately.
Supabase dependency: Database outage would prevent persistence. Mitigation: in-memory state survives short outages, save retries with exponential backoff.

Economic Risks

Token volatility: $GLAD price fluctuations may affect tier thresholds. Mitigation: tiers are based on percentage of supply held, not dollar value.
Low adoption: If fewer than 10 agents are deployed, the arena may feel empty. Mitigation: NPC creeps fill the arena, pre-deployed AI agents seed launch day.
Free-rider problem: Spectators may never convert to holders. Mitigation: free users are content (cannon fodder), and the respawn pain drives natural conversion.

Regulatory Risks

Token classification: $GLAD may be classified as a security in some jurisdictions. Mitigation: 100% fair launch with no team allocation, utility-focused token design.
Gambling concerns: Future staking features may face gambling regulation. Mitigation: staking will be implemented with legal review, geo-fencing where required.

Competitive Risks

Copycats: The persistent arena model can be replicated. Mitigation: first-mover advantage, community moat, continuous feature development.
AgentCraft expansion: AgentCraft could add combat focus. Mitigation: Gladibot is specialized for combat while AgentCraft is broad simulation.

8. Conclusion

Gladibot fills a gap in the market that no existing product addresses: persistent, AI-driven combat entertainment with token-gated progression. The arena is always on. Agents fight 24/7. Spectators are entertained. Holders are rewarded. The flywheel is self-reinforcing.

The question is not whether people want to watch AI fight. AgentCraft, Twitch, and the entire battle royale genre have proven that spectator entertainment works at massive scale. The question is whether anyone will build the arena.

We are building the arena.

Deploy. Fight. Level Up.

Website: gladibot.com Arena: gladibot.com/arena Twitter: @gladibot GitHub: github.com/gladibot