Skip to content

DOCS_INDEX.md

Latest commit

 

History

History
116 lines (88 loc) · 7.77 KB

DOCS_INDEX.md

File metadata and controls

116 lines (88 loc) · 7.77 KB

📚 CoreAI Documentation — Index

Pick the track that matches your current goal. Every link lands on a self-contained guide; you don't need to read them in order.

Manifests: com.nexoider.coreai (1.7.5) · com.nexoider.coreaiunity (1.7.5) Repo README: ../../../README.md · Changelog: ../CHANGELOG.md

🟢 Beginner — get something on screen

Install → open scene → connect LLM → press Play.

# Document You will learn
1 QUICK_START.md Install, open _mainCoreAI.unity, wire up a backend
1b QUICK_START_FULL.md Full 10-min walkthrough: LM Studio → Unity → first command
1a COREAI_SINGLETON_API.md 🎯 One class for everyone — CoreAi.AskAsync / Stream / TryGet* (beginner + pro guide)
1c README_CHAT 💬 Drop-in chat panel + demo scene in one click
2 AGENT_BUILDER 🏗️ Build an NPC in 3 lines, agent modes, recipes
3 COREAI_SETTINGS ⚙️ LLM modes, routing profiles, models, timeout, streaming toggle
4 CHAT_TOOL_CALLING 🛒 Worked example: merchant NPC with inventory
4b EXAMPLES 📖 Enemies, crafting, auto-repair, merchant, guard

💬 Chat & Streaming — new in 0.20

Document Topic
COREAI_SINGLETON_API 🎯 One-line API: CoreAi.AskAsync / StreamAsync / OrchestrateAsync
README_CHAT CoreAiChatPanel, CoreAiChatConfig, FAB/collapse, hotkeys (0.25.3+), persisted session (0.25.4+), SubmitMessageFromExternalAsync (0.25.5+), reliable Stop path (0.25.6+), streaming hierarchy; default window ~650×910, flush-right scrollbar, optional coreai-long-request-hint (long-turn status)
STREAMING_ARCHITECTURE SSE / LLMUnity → ThinkBlockStreamFilter → UI; orchestrator streaming; cancellation; known limits
STREAMING_WEBGL_TODO WebGL: UnityWebRequest vs optional WebGlNativeStreaming (fetch jslib); history + verification checklist
WEBGL_BUILD_TROUBLESHOOTING WebGL Player build: LLVM OOM during IL2CPP, IOException under ProjectSettings/Packages, StreamingAssets preprocess log
HTTP_TRANSPORT_SPEC IOpenAiHttpTransport — HttpClient vs UnityWebRequest (WebGL); simulated streaming

🟡 Intermediate — tools, memory, roles

# Document You will learn
5 TOOL_CALL_SPEC 🔧 All built-in tools: memory, Lua, world, inventory, scene, camera
5b JSON_COMMAND_FORMAT 📋 JSON command format per role (reference)
6 MemorySystem 🧠 MemoryTool vs ChatHistory, per-role config
6a MEMORY_STORE_CUSTOM_BACKENDS 💾 IAgentMemoryStore: PlayerPrefs, cloud save, composite (offline-first)
7 AI_AGENT_ROLES 🤖 5 built-in roles, model selection strategy
8 WORLD_COMMANDS 🌍 Spawn/move/scene control from sandboxed Lua
9 LLMUNITY_SETUP_AND_MODELS 📦 LLMUnity, GGUF, OpenAI HTTP, Lua pipeline
9b TROUBLESHOOTING 🔧 Model silent, Lua crashed, memory not written, PlayMode HTTP 500 / LM Studio (0.25.7+); CoreAI → Delete All Persistent Saves... clears persistentDataPath/CoreAI

🔴 Architecture — how it works inside

DI, threading, spec, pipelines.

# Document You will learn
10 DEVELOPER_GUIDE 🗺️ Code map, LLM → commands pipeline, PR checklist; child scope + GlobalMessagePipe for LLM subscribers
10a ARCHITECTURE Clean architecture layers, LLM modes, MessagePipe; § child LifetimeScope vs GlobalMessagePipe; § source comments (English, TODO/HACK)
10c CODE_AUDIT_AND_FOLLOWUPS Comment/language backlog and logic notes from manual audits
10b COMMAND_FLOW_DIAGRAM 🗺️ Diagram: how a command travels through the system
11 DGF_SPEC 📐 Normative spec: DI, threads, authority, §9.4 main-thread rules
12 MEAI_TOOL_CALLING 🛠️ MEAI pipeline: ILlmTool → AIFunction → FunctionInvokingChatClient
12a LLM_ROUTING 🔀 Portable routing modes, policy hooks, usage sinks, timeouts vs HTTP
12b MEAI_TOKENS_FACT_VS_ESTIMATE 📊 Provider usage vs pre-request estimates; SSE stream_options; HTTP vs orchestrator timeouts; tool diagnostics
12c CoreAI/Docs/README 📑 Index of every guide under Assets/CoreAI/Docs
13 MULTIPLAYER_AI 🌐 Multiplayer + AI: host authority, replication
14 GameTemplateGuides/INDEX 📚 Per-title guides: networking, orchestration, roles

🧪 Tests — documentation

Document Tests Scope
CraftingMemory_README 5 🤖 Crafting workflow: Creator → CoreMechanic → Programmer
Play Mode layout — FastNoLlm vs LlmVerification vs Scenarios (+ Shared, LlmInfra)
ThinkBlockStreamFilterEditModeTests 24 Streaming <think> filter, including split-tag cases
SecureLuaSandboxEditModeTests — Sandbox escape: io/os/debug/load/loadfile/dofile, step/timeout guard
SmartToolCallingChatClientEditModeTests — Duplicate detection, AllowDuplicates, missing tools, exceptions
InGameLlmChatServiceEditModeTests — Sliding-window rate limiter
CoreAiChatServiceEditModeTests — Streaming enablement hierarchy (UI → per-agent → global)
LuaExecutionPipelineEditModeTests 8 Lua sandbox: exec success/failure, repair loop, role isolation
MultiAgentCraftingWorkflowPlayModeTests 2 Full workflow over a live LLM
CraftingMemoryViaLlmUnityPlayModeTests 1 Local GGUF: 4 crafts + determinism
CraftingMemoryViaOpenAiPlayModeTests 2 HTTP: 4 crafts + 2 quick crafts
CoreAiChatPanelStopPlayModeTests 1 StopAgent() cancels active request CTS, clears sending/streaming
AgentMemoryWithRealModelPlayModeTests 1 Real LLM memory write + recall (Ignore on recall 5xx after retries, 0.25.7+)

🎮 Example game (Assets/_exampleGame)

Document Purpose
UNITY_SETUP Step-by-step RogueliteArena scene setup
ARENA_ARCHITECTURE_AND_AI Arena architecture for multiplayer + AI roles
README Concept, stack, folder layout
ROGUELITE_PLAYBOOK Gameplay: run loop, meta progression

🎬 Demo & media

Document Purpose
DEMO_RECORDING_GUIDE Video/GIF capture scenarios, tools, DemoRunner script

🗺️ Roadmap

Live plan and recently-found gaps: ../../../TODO.md.