AI Town Design Document
AI Town Design Document
1. Project Overview
- Type: A lightweight simulation + social + quest-based mini-game driven by multi-agent NPCs
- Core selling points: NPCs "remember you" and collaborate with each other through dialogue; players can use community contribution points (earned by posting, submitting PRs, etc.) to obtain in-game currency and abilities, driving town events
- Technology foundation: Godot 4 (Microverse-style) + multi-agent framework (O-R-P-A: Observe → Retrieve → Plan → Act) + local model first (with template fallback)
2. Goals (MVP)
- Single map + 3 NPCs (merchant / messenger / editor) + quest board (fetch / relay / check-in)
- Dialogue with short-term memory + end-of-day summary
- Minimal community integration: support entering a "redemption code" to receive coins / action points (future: automatic issuance via webhook)
- Use points (or redeemed coins) to trigger 2–3 visible world changes (discount day / extra quests / expanded dialogue budget)
3. Core Gameplay (Version 1)
- Loop: Accept quest → Dialogue / collaborate with NPCs → Complete to earn coins / AP → Nightly summary generated → Events refresh next day
- Uses of points / coins (pick 2–3 to implement first)
- Unlock a discount day at the shop (all prices -10%)
- Purchase action points (one extra quest per day)
- Purchase dialogue budget (3 additional conversation turns with an NPC that day)
- Trigger a theme-day announcement (published by the editor NPC; NPC dialogue becomes more active)
4. Open-Source Community Integration (Two Phases)
Phase A (MVP) — Redemption Code Verification
- The community backend issues one-time redemption codes (containing point value and expiry); player enters code in-game → server verifies and voids → returns coins / AP
- Advantage: No login or account binding required; maximally stable and ready to ship
Phase B (Mid-term) — Automatic Issuance via Webhook
- Posts, PR merges on GitHub / the site trigger a Webhook → write to
pending_rewards - Game launch or clicking "Sync" → fetch pending rewards → automatically credited
- Optional: bind Steam / GitHub account for stronger identity verification
5. System Architecture (Minimal Modules)
-
Client (Godot)
Wallet(authoritative entry point for coins / AP)TaskManager,DialogManager,MemoryManager,CharacterManagerTownEventBus(broadcasts shop open / midday break / close / theme day)RedeemPanel(redemption code UI)
-
Services (can be merged into community backend)
/api/v1/redeem(one-time verification and voiding)- (Reserved)
/api/v1/rewards/pending,/webhooks/github
Data Flow (MVP)
Community issues code → Player enters it in-game → redeem verifies → Returns coins / AP → Wallet credits → TownEventBus triggers discount / quest refresh
6. Scoring and Spending (Initial Draft)
| Action | Community Points Earned | In-game Conversion (Example) |
|---|---|---|
| Post approved | +80 | 80 pts = 400 coins |
| PR merged | +80 | 80 pts = 400 coins |
| Article featured | +50 | 50 pts = 1 "theme day" item |
Conversion rates are stored in a config file; events can apply temporary bonuses (e.g., 1.2× on weekends)
贡献者
最近更新
Involution Hell© 2026 byCommunityunderCC BY-NC-SA 4.0
Projects
Involution Hell's documentation hub for open-source projects initiated and maintained by community members, spanning CS/AI job-hunting tools and learning resources. Perfect for developers and job seekers looking to gain project experience, contribute to open source, or prepare for technical interviews.
Multimodal Reinforcement Learning Project (MVP Goals)
Build a lightweight multimodal understanding and generation system that closes the loop from visual perception to language expression, incorporating reinforcement learning and answer-to-image generation.