@modusensus/dsh-mneme 0.7.27 → 0.7.29

package/README.en.md

@@ -174,6 +174,7 @@ Every **background LLM call** (autoDream consolidation + summary, autoSummarize
 
 | Version | Highlights |
 |------|------|
+| **v0.7.28** | Connectivity test triple fix: the `POST /api/dsh-mneme/test-model` minimal call raised `maxTokens` 16→1024 — a thinking model's reasoning alone can exhaust a 16-token budget, leaving `reply` always empty ("actually answered ok, not just connected" never held for thinking models; billed on actual usage, manual button, no amplification cost); connectivity test state is now per-route (dream/sleep each hold their own) — previously a single shared state made both groups show "testing/result" and disable each other when either button was clicked; the panel's mount-time draft now includes `sleepProvider`/`sleepModel` — values were always saved server-side, but after a remount (tab switch / re-enter) the dropdowns showed "follow default route" as if the setting had been reset; the saved values restore without re-entering them; 733 tests green |
 | **v0.7.27** | v0.7.26 endpoint backfill: the v0.7.26 release tree did not include the `fix/dream-effort-trap` branch, so the `GET /api/dsh-mneme/llm-providers` + `POST /api/dsh-mneme/test-model` endpoints announced in its CHANGELOG/Release were absent from the package (the panel's cascading dropdown + connectivity test would 404); this release rebases and merges the branch (PR #100), landing the endpoints — panel connectivity test now works; 733 tests green |
 | **v0.7.26** | Consolidation `UNSUPPORTED_REASONING_EFFORT` root-cause fix + dream/sleep model connectivity test: root-caused the **defaultEffort trap** — the harness injects `reasoning.defaultEffort` when effort is omitted, so if that default tier itself is unsupported, retrying without effort never helps (whatever effort is sent gets rejected); fix: new `resolveDreamEffort` proactively queries `ctx.llm.resolveModelInfo()` for the model's supported effort tiers and sends a supported one — an unsupported configured tier auto-falls back to the model's default/first supported tier, and the field is omitted when the model declares no reasoning capability. Settings panel adds selection hints on 6 dream/sleep model fields (guiding toward non-thinking models so they don't burn the token budget on reasoning). New `GET /api/dsh-mneme/llm-providers` (host-side provider/model discovery; keys never touch the plugin side) + `POST /api/dsh-mneme/test-model` (connectivity test; empty body resolves via the consolidation route and returns `modelId`); 732 tests green |
 | **v0.7.25** | Tool-compat hardening + `memory_get`/render content preview + consolidation-model guide: new `memory_get` tool (8th model tool, read a single memory's full body by id) + `memory_search`/`memory_list` render now embeds title/metadata/body preview (the model reads actual memory content even when the host only forwards render text) + consolidation-model selection guide (config comments/README classification, non-thinking vs thinking). Fixed `memory_get` execute nested inside output crashing with `userExecute is not a function` (previous tests only counted tool names, never executed it) + tool de-dup on live patch reload + Standalone API port-collision retry (EADDRINUSE) + client inject declaration alignment + better-sidebar hardening; 718 tests green |
@@ -234,8 +235,12 @@ Every **background LLM call** (autoDream consolidation + summary, autoSummarize
 | v0.7.20 | ✅ Done | Heat restore + phase-two frontend + better-sidebar fix | Heat model fully restored (issue #87, backported from v0.7.10: power-law decay + TYPE_DECAY + sleep heat-combined dual protection + entity heat projection) + acceptance checklist landed (heatEnabled default OFF / feature_flags 31 keys / lightMode linkage / sleep demotion audit / updated_at⊥last_accessed_at contract) + phase-two frontend (/list heat projection, HeatBadge three-tier, order=heat page-local sort) + better-sidebar fix (issue #88: inner dynamic sub-plugin); 685 tests green |
 | v0.7.21 | ✅ Done | effort-fallback stream fix | Catch-based effort fallback was dead code on the stream path (dsh-llm rc.1 turns adapter failures into a terminal error finish chunk instead of a throw) → `streamText` captures the finish-chunk cause (`describeStreamFailure`) + `withEffortFallback` gains a `getStreamError` accessor (retries without effort when rejected) + `runAuditedLlm` supports `spec.streamError` (real cause in audit); 688 tests green |
 | v0.7.22 | ✅ Done | skipInvalid tolerant-validation restore (issue #89) + autoDream throttle | Restored v0.6.9 skipInvalid dual-track structure (lost in the v0.7.11 rewrite): `dreamSkipInvalid` skips individual invalid decisions + applies the valid subset + marks the run degraded; `allowCrossTypeMerge` explicitly relaxes cross-type merging; weak models (qwen3.8-flash) with jittery schema compliance no longer fail the whole batch. New `dreamMinIntervalMinutes` (0–10080, default 0) minimum trigger interval — failed/degraded runs also consume it. Strict mode / sleep path unchanged; feature_flags whitelist 34 keys; 696 tests green |
+| v0.7.23 | ✅ Done | "Consolidation keeps failing" root fix + empty-body part 2 + skipInvalid splice fix | A legal empty decision array `[]` no longer fails the whole run (CONSOLIDATION_PROMPT explicitly allows no output when nothing needs changing, yet `validateDecisions` hard-rejected it as non-empty — model-agnostic, ChatGPT/Claude hit the same trap; fix: empty array short-circuits to `ok:true` no-op instead of tripping the implicit-keep coverage check); `dreamMaxTokens` default 8192→32768 (thinking models burn the whole budget on reasoning); skipInvalid splice residue bug (length equality ≠ content equality, skipped decisions leaked into apply/audit); 712 tests green |
+| v0.7.24 | ✅ Done | DSH Desktop plugin-tree load crash fix (v0.7.23 regression) | cordis 4's ctx is a Proxy — accessing a property not declared in `inject` throws `cannot get property "webServer" without inject` (not `undefined`), and removing webServer from inject made Desktop crash on restart; fix: restored webServer to inject (cordis applies the plugin only after the host service is ready) + apply/register guard switched to `ctx.reflect.get` (inject-free read, returns `undefined` when absent, never throws); verified with a real cordis + dsh-host-webserver plugin (API routes 200 / unknown path 404 / headless silently inactive); 714 tests green |
 | v0.7.25 | ✅ Done | Tool-compat hardening + content preview + consolidation-model guide | New `memory_get` tool (8th model tool, full-body read by id) + `memory_search`/`memory_list` render embeds title/metadata/body preview + consolidation-model selection guide; fixed `memory_get` execute nested inside output (`userExecute is not a function`) + tool de-dup on live patch reload + Standalone API port-collision retry + client inject alignment + better-sidebar hardening; 718 tests green |
 | **v0.7.26** | ✅ Done | Consolidation effort-trap root fix + LLM connectivity test | `UNSUPPORTED_REASONING_EFFORT` root cause (defaultEffort trap: harness injects `reasoning.defaultEffort` when effort is omitted, so an unsupported default can never be retried) → `resolveDreamEffort` proactively queries `ctx.llm.resolveModelInfo()` for the model's supported effort tiers (unsupported configured tier falls back to the model's default/first supported tier; field omitted when the model declares no reasoning capability); settings panel adds selection hints on 6 dream/sleep model fields; new `GET /api/dsh-mneme/llm-providers` + `POST /api/dsh-mneme/test-model` (connectivity test; empty body resolves via the consolidation route; keys never touch the plugin side); 732 tests green |
+| **v0.7.27** | ✅ Done | v0.7.26 endpoint backfill | The v0.7.26 release tree did not include the `fix/dream-effort-trap` branch, so the `GET /api/dsh-mneme/llm-providers` + `POST /api/dsh-mneme/test-model` endpoints announced in its CHANGELOG/Release were absent from the package (the panel's cascading dropdown + connectivity test would 404); this release rebases and merges the branch (PR #100), landing the endpoints — panel connectivity test now works; 733 tests green |
+| **v0.7.28** | ✅ Done | Connectivity test triple fix | `POST /api/dsh-mneme/test-model` minimal call `maxTokens` 16→1024 (a thinking model's reasoning alone can exhaust a 16-token budget, leaving `reply` always empty); connectivity-test state now per-route (dream/sleep each hold their own — buttons no longer cross-disable); mount-time draft now includes `sleepProvider`/`sleepModel` (values were always saved server-side, but after a remount the dropdowns showed "follow default route" as if reset); 733 tests green |
 | **v0.8.0** | 🚧 Planned (late Sep) | Graph enhancement | Interest-drift visualization + scope isolation (issue #17) + cross-workspace sharing |
 
 > All new capabilities ship as **toggleable features** (enabled/disabled via configuration), conservatively on by default and never breaking existing behavior. The `failure_memories` table and the autoDream decision engine have already paved the way for future reflective growth.

package/README.md

@@ -5,7 +5,7 @@
 [![npm version](https://img.shields.io/npm/v/@modusensus/dsh-mneme?color=blue&label=npm)](https://www.npmjs.com/package/@modusensus/dsh-mneme)
 [![license](https://img.shields.io/badge/license-MIT-green)](LICENSE)
 [![Awesome](https://awesome-dsh-plugin.com/badge.svg)](https://github.com/awesome-dsh-plugin/awesome-dsh-plugin)
-[![tests](https://img.shields.io/badge/tests-714%20passed-success)](https://github.com/modusensus/dsh-mneme)
+[![tests](https://img.shields.io/badge/tests-744%20passed-success)](https://github.com/modusensus/dsh-mneme)
 [![CI](https://img.shields.io/github/actions/workflow/status/modusensus/dsh-mneme/ci.yml)](https://github.com/modusensus/dsh-mneme/actions)
 [![node](https://img.shields.io/badge/node-24%2B-blue)](https://nodejs.org)
 [![npm downloads](https://img.shields.io/npm/dm/@modusensus/dsh-mneme?color=blue&label=downloads)](https://www.npmjs.com/package/@modusensus/dsh-mneme)
@@ -215,6 +215,8 @@ v0.3.0 起新增**记忆基因**层：从记忆里抽取**命名实体**、**带
 
 | 版本 | 亮点 |
 |------|------|
+| **v0.7.29** | 实体抽取路由契约修复 + 面板控件 + 帮助与反馈：`streamEntityText` 抽取为可测试导出的 `createEntityStreamAdapter`——显式 `provider`/`model` 优先、缺省兜底读默认路由选择；实体抽取思考强度 `entityExtractionReasoning` 被模型拒绝时自动去 effort 重试（与 autoDream/sleep 同一降级策略）；设置面板新增实体抽取 `provider`/`model`/思考强度三控件；面板底部新增「帮助与反馈」卡片（GitHub issue 预填 + mailto `work@modusensus.space` + 浏览已知问题）+ 配套 `GET /api/dsh-mneme/info`；744 测试全绿 |
+| **v0.7.28** | 连通性测试三连修：`POST /api/dsh-mneme/test-model` 最小调用 `maxTokens` 16→1024——思考模型的推理过程足以耗尽 16 token 预算致 `reply` 恒为空（「真的答了 ok 而非只报通」对思考模型不成立；按实际用量计费，手动按钮无放大成本）；连通性测试状态按巩固/睡眠路由各持一份——此前共享单份，点任一「测试连通性」按钮两组同显「测试中/结果」且互相禁用；面板挂载草稿补 `sleepProvider`/`sleepModel`——后端一直保存正常，但重挂载（切页签/退出重进）后下拉恒显「跟随默认路由」，看起来像设置被重置，补键后正常回填（历史已存值无需重填）；733 测试全绿 |
 | **v0.7.27** | v0.7.26 端点遗漏补齐：v0.7.26 发版树未含 `fix/dream-effort-trap` 分支，CHANGELOG/Release 宣告的 `GET /api/dsh-mneme/llm-providers` + `POST /api/dsh-mneme/test-model` 两个端点实际不在包内（面板「级联下拉 + 测试连通性」会 404）；本版将分支 rebase 后合入（PR #100），端点落地，面板连通性测试可用；733 测试全绿 |
 | **v0.7.26** | 记忆巩固 `UNSUPPORTED_REASONING_EFFORT` 根治 + 巩固/睡眠模型连通性测试：根治 **defaultEffort 陷阱**——harness 在 effort 省略时注入 `reasoning.defaultEffort`，若该默认档位本身不被模型支持，则去 effort 重试也无济于事（换什么 effort 都会被拒绝）；修复：新增 `resolveDreamEffort` 经 `ctx.llm.resolveModelInfo()` 主动探测模型支持的档位再发送——配置档位不被支持时自动换用模型支持的默认/首个档位、模型声明无 reasoning 能力则省略字段；设置面板巩固/睡眠 6 字段补选型提示（引导非思考模型，避免再踩思考模型烧光 token 预算）；新增 `GET /api/dsh-mneme/llm-providers`（宿主侧 provider/model 发现，密钥不经插件侧）+ `POST /api/dsh-mneme/test-model`（模型连通性测试，空 body 按巩固路由解析返回 modelId）；732 测试全绿 |
 | **v0.7.25** | 工具兼容性加固 + `memory_get`/render 内容预览 + 巩固模型引导：新增 `memory_get` 工具（第 8 个模型工具，按 id 读单条记忆全文）+ `memory_search`/`memory_list` render 嵌入标题/元数据/正文预览（宿主只透传 render 文本时模型也能直接读到记忆内容）+ 巩固模型选型引导（config 注释/README 分类声明，非思考 vs 思考模型差异）；修复 `memory_get` execute 嵌套进 output 的崩溃（`userExecute is not a function`，此前测试只数工具名从未执行该工具而掩盖）+ 工具重复注册去重（live patch reload）+ Standalone API 端口冲突重试（EADDRINUSE）+ client inject 声明对齐 + better-sidebar 集成加固；718 测试全绿 |
@@ -290,9 +292,9 @@ v0.3.0 起新增**记忆基因**层：从记忆里抽取**命名实体**、**带
 | **v0.7.3** | ✅ 完成 | issue #38 新功能 | 左下角入口按钮可选开关 `showSidebarTrigger`（默认开，settings-over-config）；Web 面板设置一键关闭，与 dsh-cost-meter 等 footer 插件冲突可隐藏按钮、记忆库标签不受影响；776 测试全绿 |
 | **v0.7.4** | ✅ 完成 | issue #40 + #41 修复 | 注入边界 run-based 花括号转义（`{{a}}`→`{\{a\}\}`、奇数连续如 `{{{a}}}` 不残留字面，`escapePromptVariables` 默认开）+ 记忆窗口顶栏左对齐、关闭按钮避开宿主窗口控制按钮区；782 测试全绿 |
 | **v0.7.5** | ✅ 完成 | 分层记忆类型 + 总览视图 | 借鉴 meow-memory 分层概念、贴合单表架构：新增 `user`（用户画像）/`fact`（原子事实）类型，注入/镜像/梦境/质量过滤全链路打通；Web 面板「总览」视图（分层卡片 + 用户画像卡 + 类型分布 + 近 7 天趋势）；`/api/dsh-mneme/stats` 端点；kimi-k2.7-code 复验；790 测试全绿 |
+| **v0.7.6** | ✅ 完成 | issue #48 修复 | 四工具统一 `service.resolveMemoryId`：截断/前缀短 id 也能精确操作（精确命中优先、唯一前缀解析、多命中拒绝并列出候选、`memory_delete` 未命中幂等返回 + `logger.warn` 留痕）；Web bundle `client.js` 改 src 正源；801 测试全绿 |
 | **v0.7.7** | ✅ 完成 | issue #23 图谱回填 | sleep 批量实体抽取 phase：默认关 `sleepEntityExtractionEnabled`，按最老优先、SQL LIMIT/OFFSET 分页（下沉为有界查询，不再整表扫描）批量抽取未打标记忆的实体（修复 issue #23 实体图谱空白）；`pending_extracted_at` 幂等防重、成功/失败清除，metadata 合并不覆盖；node:sqlite 兼容修复（pluck→all+map、`forgotten=0` 查询条件）；810 测试全绿 |
 | **v0.7.8** | ✅ 完成 | DSH 0.1.2-rc.1 兼容（issues #58 #59） | 官方移除 `Session.events` 属性、改用 `snapshotEvents()` 方法后 autoSummarize 与 hot-context 注入失效；改为兼容垫片 `session.snapshotEvents?.() ?? session.events`，新旧 DSH 通吃，老版本不受影响；新增 2 个回归用例；812 测试全绿 |
-| **v0.7.6** | ✅ 完成 | issue #48 修复 | 四工具统一 `service.resolveMemoryId`：截断/前缀短 id 也能精确操作（精确命中优先、唯一前缀解析、多命中拒绝并列出候选、`memory_delete` 未命中幂等返回 + `logger.warn` 留痕）；Web bundle `client.js` 改 src 正源；801 测试全绿 |
 | **v0.7.9** | ✅ 完成 | issue #65 修复 | v0.7.8 的 snapshotEvents 适配未同步 lib/（npm 实际加载 lib/）导致发布产物静默失效；补齐 lib 三处垫片 + 发布前 src↔lib 一致性校验（`scripts/check-sync.js`，root prepack 调用，漂移 exit 1）+ `test/lib-smoke.test.js` 从 lib 导入复跑 + 一致性断言；815 测试全绿 |
 | **v0.7.10** | ✅ 完成 | Web 面板体验升级 | 记忆类型色点体系 + 图谱画布平移/缩放（0.5x–3x 光标锚定）+ 设置页分区重排 + 侧边栏同标签冲突修复 + 详情 meta 精排 + 只读端点 `GET /api/dsh-mneme/entities`；815 测试全绿 |
 | **v0.7.11** | ✅ 完成 | 记忆库面板改版 | 按月分页 + 无限滚动 + 搜索全局化 + 30s 静默刷新 + 状态子页四卡 + 删除两步确认 + 重要性过滤芯片 + 内联 Lucide 图标；bundle 默认开实体抽取；修复 issue#72 图谱节点裁剪 + issue#59 autoSummarize 垫片；595 测试全绿 |
@@ -305,11 +307,14 @@ v0.3.0 起新增**记忆基因**层：从记忆里抽取**命名实体**、**带
 | **v0.7.18** | ✅ 完成 | 生态第一步 + 查询收敛 | better-sidebar 软集成（inject 声明 + optional peer `dsh-better-sidebar` + registerTab 复用四视图，未装安全跳过；窄容器 `@container` 适配）+ `/list?deposited=only` 沉淀视图（receipt_chain ∪ source=dream）+ 记忆库沉淀/已归档筛选 chip + 状态页仪表盘化（统计 + 查看全部跳转预置筛选）+ 抽屉归档记忆「恢复」；667 测试全绿 |
 | **v0.7.20** | ✅ 完成 | heat 回归 + 阶段二前端 + better-sidebar 修复 | heat 热度模型完整找回（issue #87，v0.7.10 移植：幂律衰减 + TYPE_DECAY + sleep 热联合双保护 + 实体热投影）+ 验收清单落地（heatEnabled 默认关 / feature_flags 31 键 / lightMode 联动 / sleep 降级审计暴露 / updated_at⊥last_accessed_at 契约）+ 阶段二前端（/list heat 投影、HeatBadge 三档、order=heat 页内排序）+ better-sidebar 修复（issue #88：内层动态子插件）；685 测试全绿 |
 | **v0.7.21** | ✅ 完成 | effort 回退流式修复 | autoDream/sleep 的 catch 式 effort 回退在流式路径是死代码（dsh-llm rc.1 把 adapter 异常转成终态 error finish chunk 不再抛出）→ `streamText` 捕获 finish-chunk 失败原因（`describeStreamFailure` 归一化）+ `withEffortFallback` 增加 `getStreamError` 访问器（effort 被拒去重试）+ `runAuditedLlm` 支持 `spec.streamError`（audit 记真实原因）；688 测试全绿 |
-| **v0.7.26** | ✅ 完成 | 巩固 effort 陷阱根治 + LLM 连通性测试 | 记忆巩固 `UNSUPPORTED_REASONING_EFFORT` 根治（defaultEffort 陷阱：harness 省略 effort 时注入 `reasoning.defaultEffort`，默认档位不被模型支持则任何重试无效）→ 新增 `resolveDreamEffort` 经 `ctx.llm.resolveModelInfo()` 探测模型支持的档位（不支持的配置档位自动换用模型支持的默认/首个档位，无 reasoning 能力则省略字段）；设置面板巩固/睡眠 6 字段补选型提示（引导非思考模型）；新增 `GET /api/dsh-mneme/llm-providers`（宿主侧 provider/model 发现）+ `POST /api/dsh-mneme/test-model`（模型连通性测试，空 body 按巩固路由解析，密钥不经插件侧）；732 测试全绿 |
-| **v0.7.25** | ✅ 完成 | 工具兼容性加固 + 内容预览 + 巩固模型引导 | 新增 `memory_get` 工具（第 8 个模型工具，按 id 读单条记忆全文）+ `memory_search`/`memory_list` render 嵌入标题/元数据/正文预览（宿主只透传 render 文本时模型也能读到内容）+ 巩固模型选型引导（config 注释/README 分类声明）；修复 memory_get execute 嵌套 output 的崩溃（userExecute is not a function）+ 工具重复注册去重（live patch reload）+ Standalone API 端口冲突重试 + client inject 声明对齐 + better-sidebar 集成加固；718 测试全绿 |
-| **v0.7.24** | ✅ 完成 | 桌面端崩溃紧急修复 | v0.7.23 把 webServer 移出 inject 致 cordis Proxy 抛 `cannot get property "webServer" without inject`（未注入属性直接访问抛错而非 undefined），桌面端重启插件树加载失败；修复：恢复 webServer 到 inject（cordis 等宿主就绪再 apply）+ apply/register 守卫改 `ctx.reflect.get`（免 inject 读取、未提供返回 undefined 不抛错，未来 headless 移出 inject 也安全）；真实 cordis + dsh-host-webserver 实测 API 路由 200 / 未知路径 404 / headless 静默不激活；714 测试全绿 |
-| **v0.7.23** | ✅ 完成 | 记忆沉淀「反复失败」根治 + 空体第二段 + skipInvalid splice 修复 | consolidation 合法空数组 `[]` no-op（CONSOLIDATION_PROMPT 允许无问题无需输出；此前 validateDecisions 硬判 non-empty → 整单 failed，模型无关、ChatGPT/Claude 同样踩中；修复：空数组显式短路 ok，不再触发隐式 keep 覆盖率误判）；`dreamMaxTokens` 默认 8192→32768（思考模型推理烧光预算根治余量）；skipInvalid splice 残留 bug（长度相等≠内容一致，被跳决策残留进 apply）；712 测试全绿 |
 | **v0.7.22** | ✅ 完成 | skipInvalid 宽容校验回归（issue #89）+ autoDream 节流 | 恢复 v0.6.9 的 skipInvalid 双轨结构（v0.7.11 重写丢失）：`dreamSkipInvalid` 单条非法决策跳过 + 合法子集应用 + run 记 degraded，`allowCrossTypeMerge` 显式放宽跨类型合并；弱模型（qwen3.8-flash）决策合规抖动不再整单拒绝；新增 `dreamMinIntervalMinutes`（0-10080，默认 0=不限）最小触发间隔，失败/degraded run 也占用间隔；严格模式/sleep 路径行为不变；feature_flags 白名单 34 键；696 测试全绿 |
+| **v0.7.23** | ✅ 完成 | 记忆沉淀「反复失败」根治 + 空体第二段 + skipInvalid splice 修复 | consolidation 合法空数组 `[]` no-op（CONSOLIDATION_PROMPT 允许无问题无需输出；此前 validateDecisions 硬判 non-empty → 整单 failed，模型无关、ChatGPT/Claude 同样踩中；修复：空数组显式短路 ok，不再触发隐式 keep 覆盖率误判）；`dreamMaxTokens` 默认 8192→32768（思考模型推理烧光预算根治余量）；skipInvalid splice 残留 bug（长度相等≠内容一致，被跳决策残留进 apply）；712 测试全绿 |
+| **v0.7.24** | ✅ 完成 | 桌面端崩溃紧急修复 | v0.7.23 把 webServer 移出 inject 致 cordis Proxy 抛 `cannot get property "webServer" without inject`（未注入属性直接访问抛错而非 undefined），桌面端重启插件树加载失败；修复：恢复 webServer 到 inject（cordis 等宿主就绪再 apply）+ apply/register 守卫改 `ctx.reflect.get`（免 inject 读取、未提供返回 undefined 不抛错，未来 headless 移出 inject 也安全）；真实 cordis + dsh-host-webserver 实测 API 路由 200 / 未知路径 404 / headless 静默不激活；714 测试全绿 |
+| **v0.7.25** | ✅ 完成 | 工具兼容性加固 + 内容预览 + 巩固模型引导 | 新增 `memory_get` 工具（第 8 个模型工具，按 id 读单条记忆全文）+ `memory_search`/`memory_list` render 嵌入标题/元数据/正文预览（宿主只透传 render 文本时模型也能读到内容）+ 巩固模型选型引导（config 注释/README 分类声明）；修复 memory_get execute 嵌套 output 的崩溃（userExecute is not a function）+ 工具重复注册去重（live patch reload）+ Standalone API 端口冲突重试 + client inject 声明对齐 + better-sidebar 集成加固；718 测试全绿 |
+| **v0.7.26** | ✅ 完成 | 巩固 effort 陷阱根治 + LLM 连通性测试 | 记忆巩固 `UNSUPPORTED_REASONING_EFFORT` 根治（defaultEffort 陷阱：harness 省略 effort 时注入 `reasoning.defaultEffort`，默认档位不被模型支持则任何重试无效）→ 新增 `resolveDreamEffort` 经 `ctx.llm.resolveModelInfo()` 探测模型支持的档位（不支持的配置档位自动换用模型支持的默认/首个档位，无 reasoning 能力则省略字段）；设置面板巩固/睡眠 6 字段补选型提示（引导非思考模型）；新增 `GET /api/dsh-mneme/llm-providers`（宿主侧 provider/model 发现）+ `POST /api/dsh-mneme/test-model`（模型连通性测试，空 body 按巩固路由解析，密钥不经插件侧）；732 测试全绿 |
+| **v0.7.27** | ✅ 完成 | v0.7.26 端点遗漏补齐 | v0.7.26 发版树未含 `fix/dream-effort-trap` 分支：CHANGELOG/Release 宣告的 `GET /api/dsh-mneme/llm-providers` + `POST /api/dsh-mneme/test-model` 两个端点实际不在包内（面板「级联下拉 + 测试连通性」会 404）；本版将分支 rebase 合入（PR #100），端点落地、面板连通性测试可用；733 测试全绿 |
+| **v0.7.28** | ✅ 完成 | 连通性测试三连修 | `POST /api/dsh-mneme/test-model` 最小调用 `maxTokens` 16→1024（思考模型推理过程即可耗尽 16 token 预算，`reply` 恒空）；连通性测试状态按巩固/睡眠路由各持一份（按钮不再串扰）；面板挂载草稿补 `sleepProvider`/`sleepModel`（重挂载后睡眠路由不再显示回「跟随默认路由」，值一直有存）；733 测试全绿 |
+| **v0.7.29** | ✅ 完成 | 实体抽取路由契约修复 + 面板控件 + 反馈入口 | 实体抽取 LLM 路由契约修复（#108/#109）：`createEntityStreamAdapter` 显式 provider/model 优先 + 兜底默认路由 + effort 拒绝自动去重试；设置面板新增实体抽取 provider/model/思考强度三控件；「帮助与反馈」卡片（GitHub issue 预填 + mailto + 浏览已知问题）+ `GET /api/dsh-mneme/info`；744 测试全绿 |
 | **v0.8.0** | 🚧 计划中（9 月末） | 图谱增强 | 兴趣漂移可视化 + scope 隔离（issue #17）+ 跨 workspace 记忆共享 |
 
 > 新能力一律做成**可开关的功能**（配置启用/关闭），默认保守开启、不破坏现有行为。`failure_memories` 表与 autoDream 决策引擎已为后续反思性成长铺好路。

package/lib/api.js

@@ -161,6 +161,16 @@ export function createApi(ctx, service, settings, commands, embedder, semantic =
     }
   });
 
+  // 反馈入口预填用的插件版本（面板「帮助与反馈」卡片拉取）。读取失败（打包
+  // 环境）返回 "unknown"，链接照常可用，纯展示信息不拦截。
+  register({
+    kind: "exact",
+    path: "/api/dsh-mneme/info",
+    handler(req, res) {
+      sendJson(res, 200, { version: PACKAGE_VERSION });
+    }
+  });
+
   register({
     kind: "exact",
     path: "/api/dsh-mneme/list",
@@ -296,7 +306,10 @@ export function createApi(ctx, service, settings, commands, embedder, semantic =
         provider,
         model,
         purpose: "dsh-mneme-connectivity-test",
-        maxTokens: 16,
+        // 给思考模型留出「思考完还有正文」的空间：16 会被 reasoning 全部吃掉，
+        // reply 恒为空，「真的答了 ok」对思考模型不成立。按实际用量计费，手动
+        // 按钮无放大风险。
+        maxTokens: 1024,
         ...(reasoningEffort && reasoningEffort !== "none" ? { reasoningEffort } : {}),
         messages: [{ role: "system", content: [{ type: "text", text: "Reply with exactly one word: ok" }] }]
       })) {

package/lib/client.js

@@ -239,6 +239,12 @@ window.__ModuleLoader__.load({
         "memory.settings.apiTokenPlaceholder": "留空 = 不鉴权（默认）",
         "memory.settings.apiTokenSave": "保存 Token",
         "memory.settings.apiTokenSaved": "Token 已保存",
+        "memory.settings.feedback.title": "帮助与反馈",
+        "memory.settings.feedback.desc": "遇到问题？先看看仓库已有的 issue，或直接把情况反馈过来：",
+        "memory.settings.feedback.newIssue": "GitHub 新建 issue（自动带上环境信息）",
+        "memory.settings.feedback.email": "邮件反馈（work@modusensus.space）",
+        "memory.settings.feedback.browse": "浏览仓库已知问题",
+        "memory.settings.feedback.hint": "反馈前先搜搜是否已有相同问题，能省一份重复的 issue～",
         "memory.settings.mode.title": "运行模式",
         "memory.settings.mode.desc": "轻量模式只保留核心的记忆读写与自动注入（关闭 autoDream 巩固、实体抽取、语义搜索等高级功能），适合只想「记住偏好」的轻量使用；标准模式开启全部功能。",
         "memory.settings.mode.light": "轻量",
@@ -276,6 +282,14 @@ window.__ModuleLoader__.load({
         "memory.features.hotMemoryEnabled.hint": "最近几轮对话原文随注入携带，不写入长期记忆",
         "memory.features.entityExtractionEnabled": "实体抽取",
         "memory.features.entityExtractionEnabled.hint": "从记忆中提取人物/项目/概念，图谱随之生长",
+        "memory.features.entityExtractionProvider": "实体抽取 Provider",
+        "memory.features.entityExtractionModel": "实体抽取模型",
+        "memory.features.entityExtractionReasoning": "实体抽取思考强度",
+        "memory.features.entityExtractionReasoning.none": "跟随默认",
+        "memory.features.entityExtractionReasoning.low": "低",
+        "memory.features.entityExtractionReasoning.medium": "中",
+        "memory.features.entityExtractionReasoning.high": "高",
+        "memory.features.entityExtractionModelHint": "Provider/模型留空 = 跟随主对话模型；思考强度 none = 服务商默认",
         "memory.features.codingRetrospect": "编码记忆蒸馏",
         "memory.features.codingRetrospect.hint": "用完整转录（含工具调用与报错）提炼踩坑、约束与被否决方案",
         "memory.features.rerankEnabled": "结果重排",
@@ -528,6 +542,12 @@ window.__ModuleLoader__.load({
         "memory.settings.apiTokenPlaceholder": "Empty = no auth (default)",
         "memory.settings.apiTokenSave": "Save Token",
         "memory.settings.apiTokenSaved": "Token saved",
+        "memory.settings.feedback.title": "Help & feedback",
+        "memory.settings.feedback.desc": "Stuck? Search the existing issues, or tell us what happened:",
+        "memory.settings.feedback.newIssue": "Open a GitHub issue (environment info prefilled)",
+        "memory.settings.feedback.email": "Email feedback (work@modusensus.space)",
+        "memory.settings.feedback.browse": "Browse known issues",
+        "memory.settings.feedback.hint": "Search for an existing issue first — it saves a duplicate.",
         "memory.settings.mode.title": "Runtime mode",
         "memory.settings.mode.desc": "Light mode keeps only the core memory read/write and auto-injection (autoDream consolidation, entity extraction and semantic search are off) — for light use where you just want preferences remembered. Standard mode enables everything.",
         "memory.settings.mode.light": "Light",
@@ -565,6 +585,14 @@ window.__ModuleLoader__.load({
         "memory.features.hotMemoryEnabled.hint": "Carry the last few turns verbatim; never written to long-term memory",
         "memory.features.entityExtractionEnabled": "Entity extraction",
         "memory.features.entityExtractionEnabled.hint": "Extract people / projects / concepts so the graph grows by itself",
+        "memory.features.entityExtractionProvider": "Entity extraction provider",
+        "memory.features.entityExtractionModel": "Entity extraction model",
+        "memory.features.entityExtractionReasoning": "Entity extraction reasoning",
+        "memory.features.entityExtractionReasoning.none": "Follow default",
+        "memory.features.entityExtractionReasoning.low": "Low",
+        "memory.features.entityExtractionReasoning.medium": "Medium",
+        "memory.features.entityExtractionReasoning.high": "High",
+        "memory.features.entityExtractionModelHint": "Provider / model empty = follow the main conversation model; reasoning none = provider default",
         "memory.features.codingRetrospect": "Coding retrospection",
         "memory.features.codingRetrospect.hint": "Distill pitfalls, constraints and rejected solutions from full transcripts (tools and errors included)",
         "memory.features.rerankEnabled": "Reranking",
@@ -950,6 +978,7 @@ window.__ModuleLoader__.load({
       ".mneme-numinput:focus{border-color:var(--dsw-alias-state-business-primary);box-shadow:0 0 0 3px color-mix(in srgb,var(--dsw-alias-state-business-primary) 15%,transparent)}",
       // 字符串开关的输入框（provider/model 等）与其子块容器
       ".mneme-strinput{box-sizing:border-box;width:240px;max-width:60%;height:30px;padding:0 10px;border-radius:8px;border:1px solid var(--dsw-alias-border-l2);background:var(--dsw-alias-bg-base,transparent);color:var(--dsw-alias-label-primary);font-family:inherit;font-size:13px;outline:none;text-align:left;transition:border-color .12s,box-shadow .12s}",
+      ".mneme-feed-link{display:block;color:var(--dsw-alias-brand,var(--dsw-alias-label-primary));text-decoration:none;font-size:13px;line-height:20px;padding:4px 2px;border-radius:6px;transition:opacity .12s}.mneme-feed-link:hover{opacity:.8;text-decoration:underline}",
       ".mneme-strinput:focus{border-color:var(--dsw-alias-state-business-primary);box-shadow:0 0 0 3px color-mix(in srgb,var(--dsw-alias-state-business-primary) 15%,transparent)}",
       ".mneme-featsub{margin:2px 0 8px;padding:10px 12px;border:1px solid var(--dsw-alias-border-l1);border-radius:10px;display:flex;flex-direction:column;gap:6px}",
       ".mneme-featsub .mneme-featnum{padding:6px 0;border-bottom:none}",
@@ -1592,8 +1621,10 @@ window.__ModuleLoader__.load({
     const FEATURE_ADVANCED_BOOLS = ["hybridInject", "selectiveInjectEnabled", "adaptiveThresholdEnabled", "reflectionUpdateEnabled", "reflectionFailureTracking", "conflictFreezeEnabled", "trustEpistemicWeighting"];
     // 字符串键（blur/Enter 提交，空串合法 = 跟随默认）：巩固模型与语义
     // 检索路线。embedProvider 是枚举，用下拉单独渲染。
-    const FEATURE_STRINGS = ["dreamProvider", "dreamModel", "localEmbedModel", "ollamaBaseUrl", "ollamaModel"];
+    const FEATURE_STRINGS = ["dreamProvider", "dreamModel", "sleepProvider", "sleepModel", "entityExtractionProvider", "entityExtractionModel", "localEmbedModel", "ollamaBaseUrl", "ollamaModel"];
     const EMBED_PROVIDERS = ["openai", "local", "ollama"];
+    // 实体抽取思考强度（issue #109）：与后端 FEATURE_FLAG_ENUMS 枚举对齐。
+    const ENTITY_REASONING = ["none", "low", "medium", "high"];
 
     function FeatureRow({ name, hint, on, disabled, onToggle }) {
       return h("div", { className: "mneme-featrow" },
@@ -1623,8 +1654,10 @@ window.__ModuleLoader__.load({
       // 巩固/睡眠模型路由下拉的数据源：GET /llm-providers 探测（云端 v0.7.26+
       // 的插件侧端点）。null = 端点不可用（404/失败）→ 回退纯文本输入，不挡旧后端。
       const [routes, setRoutes] = useState(null);
-      // 连通性测试结果：{ running: true } | { ok, ms, detail }
-      const [testState, setTestState] = useState(null);
+      // 连通性测试结果（巩固/睡眠/实体抽取各一份，互不串扰）：{ running: true } | { ok, ms, detail }
+      const [dreamTest, setDreamTest] = useState(null);
+      const [sleepTest, setSleepTest] = useState(null);
+      const [entityTest, setEntityTest] = useState(null);
 
       useEffect(() => {
         let cancelled = false;
@@ -1700,7 +1733,7 @@ window.__ModuleLoader__.load({
       // 连通性测试：POST /test-model，空 provider/model = 按巩固路由解析
       // （agent 默认）。durationMs 优先后端值，缺失时客户端兜底计时；成功行
       // 附模型的实际回复（reply，后端截 100 字符）——「真的答了 ok」而非只报通。
-      const runModelTest = async (provider, model) => {
+      const runModelTest = async (provider, model, setTestState) => {
         setTestState({ running: true });
         const started = Date.now();
         try {
@@ -1733,7 +1766,7 @@ window.__ModuleLoader__.load({
       // 适配器，用户不会选到不存在的模型）+ 连通性测试。空值 = 跟随默认
       // 路由；改动即提交（与 embedProvider 下拉一致）。当前值不在枚举里时
       // 保留为额外选项，避免静默改值。
-      const routeSelects = (providerKey, modelKey) => {
+      const routeSelects = (providerKey, modelKey, testState, setTestState) => {
         const curP = strs[providerKey] ?? "";
         const curM = strs[modelKey] ?? "";
         const entries = Array.isArray(routes) ? routes : [];
@@ -1769,7 +1802,7 @@ window.__ModuleLoader__.load({
             h("button", {
               type: "button", className: "mneme-btn",
               disabled: busy || (testState && testState.running),
-              onClick: () => runModelTest(curP, curM)
+              onClick: () => runModelTest(curP, curM, setTestState)
             }, t((testState && testState.running) ? "memory.features.modelTesting" : "memory.features.modelTest")),
             testState && !testState.running && h("div", { className: "mneme-featsubhint" },
               (testState.ok ? "✓ " + t("memory.features.modelTestOk") : "✗ " + t("memory.features.modelTestFail"))
@@ -1812,7 +1845,7 @@ window.__ModuleLoader__.load({
       // 可用时用级联下拉 + 连通性测试；旧后端（端点 404）回退纯文本输入。
       const dreamSub = eff.autoDream && h("div", { className: "mneme-featsub" },
         Array.isArray(routes)
-          ? routeSelects("dreamProvider", "dreamModel")
+          ? routeSelects("dreamProvider", "dreamModel", dreamTest, setDreamTest)
           : h(react.Fragment, null, strRow("dreamProvider"), strRow("dreamModel")),
         h("div", { className: "mneme-featsubhint" }, t("memory.features.dreamModelHint"))
       );
@@ -1820,10 +1853,30 @@ window.__ModuleLoader__.load({
       // 睡眠模型：sleepModeEnabled 开着才展开（sleepProvider/sleepModel 随本版
       // 进白名单；下拉与测试复用同一 /llm-providers 数据源）。
       const sleepSub = eff.sleepModeEnabled && Array.isArray(routes) && h("div", { className: "mneme-featsub" },
-        routeSelects("sleepProvider", "sleepModel"),
+        routeSelects("sleepProvider", "sleepModel", sleepTest, setSleepTest),
         h("div", { className: "mneme-featsubhint" }, t("memory.features.sleepModelHint"))
       );
 
+      // 实体抽取路由与思考强度（issue #109）：entityExtractionEnabled 开着才
+      // 展开。provider/model 级联下拉 + 连通性测试（同巩固/睡眠），reasoning
+      // 枚举即时提交；旧后端（/llm-providers 404）回退纯文本输入。
+      const entitySub = eff.entityExtractionEnabled && h("div", { className: "mneme-featsub" },
+        Array.isArray(routes)
+          ? routeSelects("entityExtractionProvider", "entityExtractionModel", entityTest, setEntityTest)
+          : h(react.Fragment, null, strRow("entityExtractionProvider"), strRow("entityExtractionModel")),
+        h("div", { className: "mneme-featnum" },
+          h("span", { className: "mneme-featnumlabel" }, t("memory.features.entityExtractionReasoning")),
+          h("select", {
+            className: "mneme-select",
+            value: eff.entityExtractionReasoning || "none",
+            disabled: busy,
+            onChange: (e) => put({ entityExtractionReasoning: e.target.value })
+          },
+            ENTITY_REASONING.map((r) => h("option", { key: r, value: r }, t(`memory.features.entityExtractionReasoning.${r}`))))
+        ),
+        h("div", { className: "mneme-featsubhint" }, t("memory.features.entityExtractionModelHint"))
+      );
+
       if (error && !state) return h("section", { className: "mneme-set-card" },
         h("div", { className: "mneme-set-title" }, t("memory.features.title")),
         h("div", { className: "mneme-set-hint" }, error));
@@ -1840,7 +1893,7 @@ window.__ModuleLoader__.load({
               FEATURE_GROUPS.map((g) => h(react.Fragment, { key: g.key },
                 h("div", { className: "mneme-featgroup" }, t(`memory.features.${g.key}`)),
                 g.items.map(boolRow),
-                g.key === "group.enhance" && embedSub,
+                g.key === "group.enhance" && h(react.Fragment, null, embedSub, entitySub),
                 g.key === "group.dream" && h(react.Fragment, null, dreamSub, sleepSub)
               )),
               h("div", { className: "mneme-featgroup" },
@@ -1866,6 +1919,18 @@ window.__ModuleLoader__.load({
       return `${s.slice(0, 5)}${"•".repeat(10)}${s.slice(-4)}`;
     }
 
+    // 反馈预填用的平台标签：从 userAgent 提一个粗粒度 OS 名，够定位问题即可，
+    // 不引第三方解析库。浏览器/Electron webview 里都能取到。
+    const platformLabel = () => {
+      const ua = (typeof navigator !== "undefined" && navigator.userAgent) || "";
+      if (/Windows NT/.test(ua)) return "Windows";
+      if (/Mac OS X/.test(ua)) return "macOS";
+      if (/Android/.test(ua)) return "Android";
+      if (/iPhone|iPad/.test(ua)) return "iOS";
+      if (/Linux/.test(ua)) return "Linux";
+      return "Unknown";
+    };
+
     function SettingsContent({ t }) {
       const [profile, setProfile] = react.useState("");
       const [rules, setRules] = react.useState([]);
@@ -1882,6 +1947,8 @@ window.__ModuleLoader__.load({
         (typeof window !== "undefined" && window.localStorage) ? window.localStorage.getItem("dsh-mneme-api-token") || "" : ""
       );
       const [apiTokenSaved, setApiTokenSaved] = react.useState(false);
+      // 反馈入口的插件版本（GET /info）。失败保持 "unknown"，链接仍可用。
+      const [pkgVersion, setPkgVersion] = react.useState("unknown");
       // 运行模式 — light vs standard; null = still loading. The card keeps
       // its own busy/saved/error state so it never blocks the others.
       const [mode, setMode] = react.useState(null);
@@ -1917,6 +1984,27 @@ window.__ModuleLoader__.load({
 
       react.useEffect(() => { load(); }, [load]);
 
+      // 帮助与反馈卡片的版本号：独立小请求，失败不影响其它卡片。
+      react.useEffect(() => {
+        let cancelled = false;
+        apiFetch("/api/dsh-mneme/info")
+          .then((res) => (res.ok ? res.json() : null))
+          .then((j) => { if (!cancelled && j && typeof j.version === "string") setPkgVersion(j.version); })
+          .catch(() => {});
+        return () => { cancelled = true; };
+      }, []);
+
+      // 反馈链接预填：环境信息（插件版本 + 平台）+ 问题描述骨架。邮件主题带
+      // 版本/平台方便归档，正文同款模板；纯链接零后端成本。
+      const feedbackEnv = `**插件版本**: ${pkgVersion}\n**平台**: ${platformLabel()}\n`;
+      const feedbackBody = feedbackEnv + "**问题描述**:\n- 期望行为:\n- 实际行为:\n- 复现步骤:\n";
+      const issueHref = "https://github.com/modusensus/dsh-mneme/issues/new?title="
+        + encodeURIComponent("[dsh-mneme] 问题反馈")
+        + "&body=" + encodeURIComponent(feedbackBody);
+      const mailHref = "mailto:work@modusensus.space?subject="
+        + encodeURIComponent(`[dsh-mneme 反馈] ${pkgVersion} / ${platformLabel()}`)
+        + "&body=" + encodeURIComponent(feedbackBody);
+
       // Runtime mode + external API — two independent fetches: one failing
       // endpoint only errors its own card, never the other one.
       react.useEffect(() => {
@@ -2317,6 +2405,19 @@ window.__ModuleLoader__.load({
             h("button", { className: "mneme-btn", onClick: saveToken }, t("memory.settings.apiTokenSave"))
           ),
           apiTokenSaved && h("div", { style: { marginTop: 8 } }, h("span", { className: "mneme-saved" }, t("memory.settings.apiTokenSaved")))
+        ),
+        // 帮助与反馈 — 反馈问题三入口（新建 issue 预填 / 邮件 / 浏览已知问题）。
+        // 纯前端链接零后端成本；插件版本来自 /info，平台取 userAgent。GitHub
+        // 仓库当前没有 issue 模板，故用 issues/new?title=&body= 直接预填。
+        h("section", { className: "mneme-set-sec" },
+          h("div", { className: "mneme-set-title" }, t("memory.settings.feedback.title")),
+          h("div", { className: "mneme-set-desc" }, t("memory.settings.feedback.desc")),
+          h("div", { style: { display: "flex", flexDirection: "column", gap: 2 } },
+            h("a", { className: "mneme-feed-link", href: issueHref, target: "_blank", rel: "noopener noreferrer" }, t("memory.settings.feedback.newIssue")),
+            h("a", { className: "mneme-feed-link", href: mailHref }, t("memory.settings.feedback.email")),
+            h("a", { className: "mneme-feed-link", href: "https://github.com/modusensus/dsh-mneme/issues", target: "_blank", rel: "noopener noreferrer" }, t("memory.settings.feedback.browse"))
+          ),
+          h("div", { className: "mneme-featsubhint", style: { marginTop: 8 } }, t("memory.settings.feedback.hint"))
         )
       );
     }

package/lib/config.js

@@ -178,6 +178,19 @@ export const Config = z.object({
   searchSemanticDedup: z.boolean().default(false),
   searchSemanticDedupThreshold: z.number().min(0.5).max(1).default(0.95),
 
+  // Recall fusion recipe (plan #1): how the keyword/vector/BM25 ranked lists
+  // are combined into the final ranking. `blend` (default) is the legacy
+  // behavior — weighted sum for vector/hybrid, union backfill for auto —
+  // unchanged. `rrf` (Reciprocal Rank Fusion) and `minmax` (min-max normalized
+  // weighted sum) are rank/scale-aware recipes that fix the unit mismatch the
+  // issue describes (raw cosine vs keyword score vs normalized IDF are added
+  // directly). Off by default so existing behavior holds exactly.
+  recallFusion: z.union([z.const("blend"), z.const("rrf"), z.const("minmax")]).default("blend"),
+  // Attach a `signals` object { keyword, vector, bm25, final } to each search
+  // result for transparency/debugging (plan #2). Default off; when on it only
+  // decorates the returned rows, never changes the ranking.
+  signalTransparency: z.boolean().default(false),
+
   // --- semantic: rerank layer (v0.2) --------------------------------------
   // Opt-in by default (item ⑥): the local cross-encoder pulls in onnxruntime
   // (transformers.js) at init, so a bare install must not load it. Only an
@@ -207,9 +220,23 @@ export const Config = z.object({
   // The storage layer (entities/entity_attrs/entity_relations tables + CRUD)
   // is always available regardless of this flag.
   entityExtractionEnabled: z.boolean().default(false),
+  // Optional provider override for entity extraction; empty = use the caller's
+  // default provider/model. Combined with entityExtractionModel — provider
+  // without model (or vice versa) falls through to the caller default.
+  entityExtractionProvider: z.string().default(""),
   // Optional model override for entity extraction; empty = use the caller's
   // default provider/model.
   entityExtractionModel: z.string().default(""),
+  // Reasoning effort for entity extraction (issue #109), mirrors
+  // dreamReasoningEffort: 'none' (default) omits the field / provider default;
+  // low/medium/high passed through. A provider that rejects the effort retries
+  // once without it, so opting in is safe to experiment with.
+  entityExtractionReasoning: z.union([
+    z.const("low"),
+    z.const("medium"),
+    z.const("high"),
+    z.const("none")
+  ]).default("none"),
   // Cap on entities per extraction pass and attributes per entity.
   entityExtractionMaxEntities: z.natural().min(1).max(20).default(10),
   entityExtractionMaxAttrs: z.natural().min(1).max(50).default(20),

package/lib/entities/extractor.js

@@ -146,16 +146,23 @@ export async function extractEntities(memory, { store, config, callLLM, logger }
       return { ok: false, error: "Invalid memory: missing content" };
     }
     
-    const model = config.entityExtractionModel || null;
     const systemPrompt = buildSystemPrompt(config);
     const userText = buildUserMessage(memory.content);
-    
+
     const messages = [
       { role: "system", content: [{ type: "text", text: systemPrompt }] },
       { role: "user", content: [{ type: "text", text: userText }] }
     ];
-    
-    const options = model ? { model } : {};
+
+    // Issue #109: optional provider/model override + reasoning effort are
+    // passed through to callLLM's options; the index.js adapter maps them onto
+    // the route (or falls back to the caller's default model). Empty provider/
+    // model both mean "use the caller default".
+    const options = {};
+    if (config.entityExtractionProvider) options.provider = config.entityExtractionProvider;
+    if (config.entityExtractionModel) options.model = config.entityExtractionModel;
+    const reasoning = config.entityExtractionReasoning;
+    if (reasoning && reasoning !== "none") options.reasoningEffort = reasoning;
     const llmResponse = await callLLM(messages, options);
     
     if (!llmResponse) {
@@ -219,7 +226,7 @@ export async function extractEntities(memory, { store, config, callLLM, logger }
           to_entity: toId,
           relation_type: rel.type,
           memory_id: memory.id,
-          metadata: { model: model || "default" }
+          metadata: { model: options.model || "default" }
         });
         savedRelations.push(saved);
       } catch (err) {

package/lib/index.js

@@ -35,6 +35,57 @@ export { Config };
 // value, so a `function apply` disposer would never run on unload. An arrow
 // has no prototype, is called normally, and its returned disposer is collected
 // and run by the fiber on unload.
+// Entity-extraction LLM adapter (issue #108/#109): maps the extractor's
+// options (provider/model override + reasoningEffort) onto a real dsh-llm
+// stream route and retries once without the effort when the first attempt is
+// rejected. Extracted from apply() so the effort-fallback branch is
+// unit-testable; the extractor only ever sees a callLLM(messages, options)
+// => Promise<string>. The route always carries a real provider/model (dsh-llm
+// GenerateOptions requires both) — never a bare stream.
+export function createEntityStreamAdapter({ llm, agentDefaultModel, logger }) {
+  return async function streamEntityText(messages, options = {}) {
+    let route = {};
+    if (options.provider) route.provider = options.provider;
+    if (options.model) route.model = options.model;
+    if (!route.provider || !route.model) {
+      try {
+        const sel = agentDefaultModel?.currentSelection?.();
+        if (sel?.provider && sel?.model) {
+          route.provider ??= sel.provider;
+          route.model ??= sel.model;
+        }
+      } catch { /* fall through to whatever route we already have */ }
+    }
+    const effort = options.reasoningEffort;
+    const tryStream = (withEffort) => {
+      let text = "";
+      return (async () => {
+        for await (const chunk of llm.stream({
+          ...route,
+          maxTokens: 4096,
+          ...(withEffort && effort ? { reasoningEffort: effort } : {}),
+          messages
+        })) {
+          if (chunk.type === "text-delta" && typeof chunk.text === "string") text += chunk.text;
+          if (chunk.type === "finish" && (chunk.reason?.kind === "error" || chunk.reason?.kind === "aborted")) return undefined;
+        }
+        return text;
+      })().catch((err) => {
+        logger?.warn?.(`[dsh-mneme] entity extraction llm stream failed: ${String(err)}`);
+        return undefined;
+      });
+    };
+    let text = await tryStream(true);
+    if (text === undefined && effort) {
+      // Mirror dream's effort fallback: a provider rejecting the reasoning
+      // effort must not sink the whole extraction — retry once without it.
+      logger?.warn?.(`[dsh-mneme] entity extraction: reasoningEffort "${effort}" rejected, retrying without it`);
+      text = await tryStream(false);
+    }
+    return text;
+  };
+}
+
 export const apply = (ctx, config) => {
   const rawCfg = Config(config);
 
@@ -320,36 +371,27 @@ export const apply = (ctx, config) => {
   // expects, reusing the same ctx.llm.stream consumption pattern as dream.js.
   // Explicit opt-in only (entityExtractionEnabled defaults to false); any LLM
   // failure degrades inside the extractor to { ok:false }, never a write error.
-  if (cfg.entityExtractionEnabled && ctx.llm) {
-    const streamEntityText = async (messages, options = {}) => {
-      let route = null;
-      if (options.model) {
-        route = { model: options.model };
-      } else {
-        try {
-          const sel = ctx.agentDefaultModel?.currentSelection?.();
-          if (sel?.provider && sel?.model) route = sel;
-        } catch { /* fall through to no route */ }
-      }
-      let text = "";
-      for await (const chunk of ctx.llm.stream({
-        ...(route ?? {}),
-        purpose: "entity-extract",
-        maxTokens: 4096,
-        messages
-      })) {
-        if (chunk.type === "text-delta" && typeof chunk.text === "string") text += chunk.text;
-        if (chunk.type === "finish" && (chunk.reason?.kind === "error" || chunk.reason?.kind === "aborted")) return undefined;
-      }
-      return text;
-    };
-    service.setEntityExtractor((memory) =>
-      extractEntities(memory, { store, config: cfg, callLLM: streamEntityText, logger: ctx.logger })
-        .catch((err) => {
-          ctx.logger?.warn?.(`[dsh-mneme] entity extraction failed: ${String(err)}`);
-          return { ok: false, error: String(err) };
-        })
-    );
+  if (cfg.entityExtractionEnabled) {
+    if (!ctx.llm) {
+      // Issue #108: an enabled-but-unwired extractor failed silently before —
+      // zero entities, zero llm_audit_logs, no log line anywhere. Make the
+      // missing dependency visible so a user can tell "extractor not installed"
+      // from "extraction failed".
+      ctx.logger?.warn?.("[dsh-mneme] entityExtractionEnabled=true but ctx.llm unavailable — entity extractor NOT installed");
+    } else {
+      const streamEntityText = createEntityStreamAdapter({
+        llm: ctx.llm,
+        agentDefaultModel: ctx.agentDefaultModel,
+        logger: ctx.logger
+      });
+      service.setEntityExtractor((memory) =>
+        extractEntities(memory, { store, config: cfg, callLLM: streamEntityText, logger: ctx.logger })
+          .catch((err) => {
+            ctx.logger?.warn?.(`[dsh-mneme] entity extraction failed: ${String(err)}`);
+            return { ok: false, error: String(err) };
+          })
+      );
+    }
   }
 
   const disposers = [];