SporeAI 2.9.2-beta

This update builds on the dialogue archives and context compaction introduced in 2.9.1-beta. It adds hierarchical summary merging, includes summaries in complete dialogue exports, and fixes a NeoForge startup crash caused by duplicate payload registration.

Supported versions:

• Minecraft 1.21.1 / NeoForge / Java 21
• Minecraft 1.20.1 / Forge / Java 17

New: Hierarchical Summary Merging

• Older first-level summaries can now be combined into second-level, third-level, and higher summaries, allowing older experiences to remain available for later conversations.
• Summaries remain separated by player, Mind, and memory revision. Minds with identical names are still handled independently.
• Exact duplicates are removed based on speaker, category, and text. When the remaining entries fit without further compression, they are merged locally with no model request and no AI token usage.
• When further compression is needed, merging uses the model currently configured for that Mind.
• Original dialogue, first-level summaries, and lower-level summary files are retained. Higher-level files record their sources and coverage.
• The latest 16 first-level summaries remain available separately. Older summaries are eligible for merging.
• The default group size is 8, with 2–8 summaries selected according to the input limit. The default maximum level is 4, configurable from 2 to 8.

Token Usage and Request Controls

• First-level compaction and higher-level merging share the existing cooldowns, hourly request limits, and token budgets.
• Each maintenance job makes at most one model request. Merging does not receive an additional request allowance.
• Model input focuses on deduplicated summary entries and short source indices, rather than the full personality script, creature catalog, or original transcript.
• Existing input limits remain in effect, with model output capped at 384 tokens.
• Scheduling accounts for new dialogue waiting to be summarized, reducing repeated processing of older summaries.
• The same source group will not trigger another model request within one hour. This record survives restarts, and successfully completed groups are not submitted again.
• Local merging uses no AI tokens. Model-based merging does use tokens. Actual costs depend on the model, dialogue volume, and configuration.

New settings under [AI] in the server-side spore_ai-common.toml:

memoryMergeEnabled = true
memoryMergeGroup = 8
memoryMergeMaxLevel = 4

Merging follows the existing memoryCompactionEnabled master switch. Automatic compaction remains disabled by default. Disabling memoryMergeEnabled stops the creation of new higher-level summaries; existing summaries remain stored and available for retrieval.

Retrieval and Management UI

• Existing long-term memory keeps priority, with additional context selected from recent first-level summaries and older higher-level summaries.
• Retrieval stays within the existing summary text budget rather than increasing the amount added to each request.
• Strengthened category validation for facts, promises, unfinished matters, and uncertain information, while retaining speaker and source information.
• The compaction management screen now shows merged-file counts, local-merge counts, the highest summary level, and read errors.
• New interface text is provided in Simplified Chinese, English, Russian, Japanese, French, and German.
• If higher-level summaries fail validation, existing memory and first-level summaries remain available. Normal retrieval continues to respect the current memory revision.

Complete Dialogue Exports Now Include Summaries

Open:

Personality Script → Memory Management → Dialogue Archive → Export All Archived Dialogue

• Exports original dialogue, first-level summaries, and higher-level merged summaries.
• Includes TXT, JSONL, summary JSON files, archive volumes, and manifest.json.
• The manifest records summary counts, file checksums, and summary warnings.
• Exports use a fixed cutoff. New dialogue or summaries created afterward are left for a later export.
• Invalid summaries or directories are recorded as warnings while the original dialogue continues to export.
• If a captured file changes, disappears, or causes the total size limit to be exceeded, the task stops and the original data is preserved.
• Exporting reads local files and does not call an AI model. Keep all archive volumes together with the manifest.

Fix: NeoForge Startup Crash

• Fixed a crash before the main menu with the following error:

Cannot register payload spore_ai:mind_memory_frame as it is already registered.

• Corrected duplicate registration in the memory management, dialogue export, and context compaction protocols.
• Preserved existing directional handlers, server-side permission checks, and payload size limits.
• No changes to personality data, memory file formats, or save formats.
• The Forge transport was inspected and is not affected by this duplicate-registration issue.

Updating and Validation

• Back up your saves and configuration before updating. Replace the old main mod JAR with the correct loader version, and keep client and server versions matched.
• Do not leave multiple versions of the SporeAI main mod installed.
• The appearance addon is unchanged. Continue using 1.0.0-beta.2 for your loader.

SporeAI 2.9.2-beta

本次更新在 2.9.1-beta 的对话归档与上下文压缩基础上，加入多级摘要合并，补齐摘要导出，并修复 NeoForge 的网络包重复注册启动崩溃。

支持：

• Minecraft 1.21.1 / NeoForge / Java 21
• Minecraft 1.20.1 / Forge / Java 17

新增：多级摘要合并

• 较早的一级摘要可以继续整理成二级、三级及更高层摘要，让较久以前的经历参与后续对话检索。
• 按玩家、心智与记忆版本分别处理。同名心智仍独立管理。
• 先按说话方、内容类别和正文精确去重。能够完整保留的内容在本地合并，不调用模型、不消耗 AI Token。
• 需要进一步压缩时，使用对应心智当前配置的模型。
• 保留原始对话、一级摘要和下级摘要，高层文件记录来源及覆盖范围。
• 默认保留最近 16 份一级摘要独立使用，较旧内容再分组整理。
• 默认分组数量为 8，实际每组根据输入容量选择 2—8 份；默认最高生成第四级，可调整为第二至第八级。

Token 与请求控制

• 一级摘要与多级合并共用原有冷却、每小时请求次数及 Token 预算。
• 每个整理任务最多发起一次模型请求，不额外增加调用额度。
• 模型输入以去重后的摘要条目和简短来源索引为主，不附带整套人格剧本、生物图鉴或完整原始对话。
• 沿用原有输入容量限制，模型输出最多 384 Token。
• 有新的原文积压时，兼顾原文整理，减少反复合并旧摘要。
• 同一组来源在一小时内不会重复请求模型；重启后继续保留该记录。成功完成的组合不会重复提交。
• 本地合并免费；模型合并会消耗 Token。 实际费用与模型、对话量及配置有关。

服务端 spore_ai-common.toml 的 [AI] 分类新增：

memoryMergeEnabled = true
memoryMergeGroup = 8
memoryMergeMaxLevel = 4

多级合并跟随 memoryCompactionEnabled 总开关，自动压缩默认仍为关闭。关闭 memoryMergeEnabled 只停止生成新的高层摘要，已有摘要继续保存并参与检索。

检索与管理界面

• 继续优先使用现有长期记忆，再从近期一级摘要和较早的高层摘要中补充内容。
• 沿用原有摘要文字预算，不扩大每次请求中的补充上限。
• 加强事实、约定、未完成事项和不确定信息的分类校验，保留说话方与来源信息。
• 压缩管理页新增合并文件数量、本地合并数量、最高层级及读取异常提示。
• 新增界面文字同步简体中文、英语、俄语、日语、法语和德语。
• 高层摘要异常时，继续使用现有记忆与一级摘要；日常检索保持当前记忆版本隔离。

完整对话导出现在包含摘要

入口：

人格剧本 → 记忆管理 → 对话档案 → 导出全部已归档对话

• 导出原始对话、一级摘要及高层合并摘要。
• 保留 TXT、JSONL、摘要 JSON、分卷文件和 manifest.json。
• 清单记录摘要数量、文件校验值和摘要警告。
• 采用固定截止位置，导出开始后的新内容留待下次导出。
• 摘要损坏或目录检查失败时，原文继续导出，并记录未收录原因。
• 已捕获文件随后发生变化、被删除或总大小超限时，停止任务并保留原始资料。
• 导出只读取本地文件，不调用 AI。请保存全部分卷及清单。

修复：NeoForge 启动崩溃

• 修复进入主菜单前出现以下错误的问题：

Cannot register payload spore_ai:mind_memory_frame as it is already registered.

• 同步修正记忆管理、对话导出及上下文压缩三个协议的重复注册。
• 保留原有收发逻辑、服务端权限与数据大小限制。
• 不修改人格、记忆文件或存档格式。
• Forge 的网络注册实现经检查不受此次问题影响。

更新与验证

• 更新前备份存档和配置，用对应加载器的新 JAR 替换旧主模组，客户端与服务器使用匹配版本。
• 不要同时安装多个 SporeAI 主模组版本。
• 外观附属本次未更新，继续使用对应加载器的 1.0.0-beta.2。