SSagaSmithAI
GitHub ↗
ACTIVE ALPHA · OPEN PLATFORM

让世界
记住每一次选择
Make every world
remember the choice

SagaSmithAI 是 AI 原生 TTRPG 平台。它把 Agent、可执行规则、分支时间线、角色知识与用户拥有的规则书和模组,连接成一条可审计的长期战役运行链。

SagaSmithAI is an AI-native TTRPG platform connecting agents, executable rules, branching timelines, actor-scoped knowledge, and user-owned books into one auditable campaign runtime.

RULES THAT EXECUTEWORLDS THAT BRANCHKNOWLEDGE THAT STAYS SCOPED
CAMPAIGN RUNTIME / LIVESESSION 07
01LOBBYrules · module · party
02PLAYscene · knowledge · checks
03COMBATturns · actions · temp map
CURRENT SCENE

钟楼下的密室

The Chamber Beneath the Bell

scope / party68%
ACTOR KNOWLEDGE
V

VARIS
knows: broken sigil

03
M

MIRA
knows: hidden patron

02
GM

KEEPER VIEW
sealed facts

09
21:14:06exposure

play.scene + play.resolution loaded

OK21:14:11branch

ancestry verified / ash-vault

OK21:14:18ruling

line of sight requires GM decision

WAIT
01MCP 服务端暴露Server-side MCP exposure
02分支感知连续性Branch-aware continuity
03PC / NPC 独立认知Independent PC / NPC knowledge
04可审计规则结算Auditable rule settlement
01 / PLATFORM

不只是会说话的角色卡More than a talking character sheet

AI 原生不是给旧工具加一个聊天框,而是让 Agent 的每次判断都能找到证据、受到边界约束,并把结果写回正确的世界与角色。

AI-native does not mean bolting chat onto an old tool. Every agent decision should find evidence, respect boundaries, and write back to the right world and actor.

A

规则可执行Executable rules

确定的输入由引擎结算:骰式、AC、资源、伤害类型、回合与状态。目标、视线、例外和叙事代价仍由 GM 裁决。

Engines settle dice, AC, resources, typed damage, turns, and state. Targets, sight lines, exceptions, and narrative cost remain GM rulings.

PRE-FLIGHT → CHOICE → COMMIT
B

世界会分支Worlds can branch

Snapshot DAG、branch checkout 与 revision ancestry 决定哪段历史有效。向量检索只寻找候选,绝不决定时间线真相。

Snapshot DAGs, branch checkout, and revision ancestry determine valid history. Vector retrieval finds candidates; it never decides timeline truth.

SNAPSHOT · BRANCH · CONTINUITY
C

角色各有所知Actors know differently

每个 PC、NPC 和玩家视角都独立维护。服务端按 principal、role、actor、scope 与 branch 过滤,不靠提示词保密。

Every PC, NPC, and player view is independent. The server filters by principal, role, actor, scope, and branch—not prompt etiquette.

PRINCIPAL · ACTOR · VISIBILITY
D

内容能落地Content becomes runtime

用户规则书和模组经过 staging、解析、质量报告、索引和规则包校验,保留页码、版本与来源,再进入战役。

User-owned books and adventures pass through staging, parsing, quality reports, indexing, and rule-pack validation with provenance intact.

SOURCE · PARSE · VALIDATE · PIN
02 / CAMPAIGN FLOW

从开团前到最后一骰From the lobby to the last roll

工具不是永久堆在模型上下文里。D&D MCP 根据真实战役阶段,为每个会话开放不同的能力面。

Tools do not remain piled into model context. D&D MCP exposes a different capability surface to every session based on authoritative campaign phase.

01LOBBYPREPARE

把桌子准备好Prepare the table

  • 导入规则书与模组
  • 车卡、装备与准备法术
  • 权限、分支与初始知识
  • Import books and adventures
  • Build characters and prepare spells
  • Set access, branches, and knowledge
lobby.rules · lobby.characters
02PLAYEXPLORE

让世界回应选择Let the world answer

  • 当前场景与连续性上下文
  • 调查、社交与非战斗检定
  • 事件、记忆与 actor knowledge
  • Current scene and continuity
  • Investigation, social play, checks
  • Events, memory, actor knowledge
play.scene · play.resolution
03COMBATRESOLVE

明确后再结算Resolve after clarity

  • 临时战斗地图与可见状态
  • 攻击预检、反应与选择窗口
  • 回合、资源、伤害与 Snapshot
  • Temporary map and visible state
  • Attack preflight, reactions, choices
  • Turns, resources, damage, snapshots
combat.observe · combat.actions
SESSION-SCOPED

结束战斗后,combat 工具自动失效并回到 play。不同频道、玩家与战役可以在同一个 MCP 进程上拥有不同 exposure。

When combat ends, combat tools expire and the campaign returns to play. Different channels, players, and campaigns can hold different exposures on one MCP process.

03 / ARCHITECTURE

边界清楚,Agent 才可靠Reliable agents need clear boundaries

每一层只拥有它真正应该拥有的状态。这样规则升级、系统扩展、更换 Agent Host 或多人接入时,不需要复制领域逻辑。

Each layer owns only the state it should. Rules can evolve, systems can expand, and Agent Hosts can change without copying domain logic.

01
HUMAN SURFACES

Players · GM · Channels · UI

聊天、控制台与桌面体验Conversation, controls, and table experience

trusted principal + session
02
AGENT HOST

SagaSmith Agent

身份、会话、模型、workspace 与多渠道Identity, sessions, models, workspace, and channels

MCP / tools · resources · prompts
03
CAPABILITY PLANE

Domain MCP Servers

渐进发现、会话暴露、权限、幂等与审计Discovery, session exposure, access, idempotency, audit

validated domain operations
04A
RULE SYSTEMS

D&D 5e · CoC 7e · Next

规则包、角色 schema、战斗与解析 profileRule packs, character schemas, combat, parsers

04B
SHARED CORE

Campaign Runtime

分支、知识、导入、检索与持久化Branches, knowledge, imports, retrieval, persistence

SKILLS ≠ ENGINE流程不冒充结算Procedure never impersonates settlement
SEARCH ≠ TRUTH检索不决定状态Retrieval never owns state
AGENT ≠ DATABASEHost 不拥有领域数据Hosts do not own domain data
04 / OPEN ECOSYSTEM

选择你的入口Choose your entry point

这是一个多仓库平台。每个仓库独立版本与测试,README 明确其所有权、依赖和成熟度。

This is a multi-repository platform. Every repository versions and tests independently, with explicit ownership, dependencies, and maturity.

HOSTALPHA

SagaSmith-agent

身份、会话、多渠道与 MCP 编排

Identity, sessions, channels, and MCP orchestration

OPEN REPOSITORY
MCPREFERENCE

SagaSmith-dnd-mcp

D&D 能力服务与会话级工具暴露

D&D capability service and session exposure

OPEN REPOSITORY
MCPTESTABLE

SagaSmith-coc-mcp

CoC 能力服务、角色知识与会话级工具暴露

CoC capability service, actor knowledge, and session exposure

OPEN REPOSITORY
COREACTIVE

sagasmith-core

分支、知识、导入、规则包与持久化

Branches, knowledge, imports, rule packs, persistence

OPEN REPOSITORY
RULESACTIVE

sagasmith-dnd

D&D 5e 2014/2024 规则与结构化战斗

D&D 5e 2014/2024 rules and structured combat

OPEN REPOSITORY
RULESACTIVE

sagasmith-coc

CoC 7e 检定、SAN、追逐与模组解析

CoC 7e checks, sanity, chases, and scenarios

OPEN REPOSITORY
SKILLSMCP-FIRST

SagaSmith-dnd-skills

D&D 主持、战役管理与参考资料

D&D GM operations, campaigns, and references

OPEN REPOSITORY
SKILLSCLI-FULL

SagaSmith-coc-skills

CoC 守秘、调查团与场景工作流

CoC Keeper, campaign, and scene workflows

OPEN REPOSITORY
CREATESKILL

SagaSmith-module-gen-skills

25 种范式的结构化冒险生成

Structured adventure design across 25 paradigms

OPEN REPOSITORY
UITESTABLE

sagasmith-dnd-ui

Scene Atlas、空间证据与临时战斗地图

Scene Atlas, spatial evidence, and temporary combat maps

OPEN REPOSITORY
UIPROTOTYPE

sagasmith-coc-ui

CoC Keeper 客户端探索

CoC Keeper client exploration

OPEN REPOSITORY
UIPROTOTYPE

sagasmith-ui

跨系统客户端探索

Cross-system client exploration

OPEN REPOSITORY
05 / STATUS

先打通一条真实路径Prove one real path first

D&D 是当前端到端参考实现:从 Agent 会话 exposure、规则书/模组导入、角色与知识、Scene Atlas、非战斗到临时战斗地图和 Snapshot。CoC 已打通独立 MCP、玩家角色授权、调查模组、SAN/战斗/追逐判定与持久化重启链路。

D&D remains the end-to-end reference, including the principal-aware UI gateway and combat map. CoC now has an independent MCP vertical slice covering player/actor authorization, scenarios, sanity/combat/chase resolution, and restart-safe persistence.

阅读平台总览Read the platform overview
D&D MCP PATH
REFERENCE
CORE RUNTIME
ACTIVE
CoC MCP PATH
TESTABLE
CLIENT UI
INTEGRATED ALPHA
进度条表达相对成熟度,不是发布完成百分比。Bars show relative maturity, not release-completion percentages.
06 / FIELD NOTES

锻造日志Forge notes

架构调整、实测路径与重要边界的简短记录。

Short notes on architecture, real test paths, and important boundaries.

D&D Workbench 打通 Scene Atlas 与临时战斗地图

D&D UI 现在区分三类数据:按模组顺序组织的 Scene Index、只表达来源证据的 Scene Spatial,以及战斗开始时创建的临时五尺格 Combat Map。背景占位素材不携带墙体、掩体或视线等机械含义。 本地 principal-aware Gateway 通过 MCP 工具投影场景、进度与战斗状态,并用 SSE 通知 campaign revision 变化。拖动 Token 只提出移动请求,最终仍由 MCP 验证…

SagaSmithAI 转向 AI 原生 TTRPG 平台

品牌与仓库信息架构统一为 AI-native TTRPG platform。D&D 参考路径现在明确为 Agent → MCP session exposure → rules/runtime → Core,并以 lobby、play、combat 三阶段控制工具发现与调用。 官网、组织 Profile 和全部仓库 README 同步标注真实边界与成熟度;D&D UI 重新定位为面向开团现场的 Alpha DM Workbench。

"CoC 独立 MCP 垂直链路可实测"

SagaSmith-coc-mcp 让 CoC Agent 不再依赖 CLI 拼接或复制 D&D 工具表。真实 stdio 客户端测试验证了目录级发现、按 session 加载和 exposure_call fallback;多人测试验证了 PC 知识互相隔离,并在重启后保留战役、模组、Snapshot 与知识状态。 SagaSmith-coc-mcp removes the need for CoC agents to compos…

S
ROLL WITH CONSEQUENCE

下一场战役,
让世界真的记得。
For the next campaign,
build a world that remembers.