memorix 1.9.9 → 1.9.11

package/CHANGELOG.md

@@ -6,6 +6,48 @@ All notable changes to this project will be documented in this file.
 
 No unreleased changes.
 
+## [1.9.11] - 2026-10-16
+
+### Fixed
+- **CJK full-text search** -- the in-memory Orama index now keeps the default
+  Latin tokenizer behavior while adding bounded character n-grams for Chinese,
+  Japanese, and Korean text, so exact titles and short CJK queries work when
+  embeddings are disabled.
+- **Runtime supply-chain baseline** -- refreshed the MCP SDK, Git adapter,
+  HTTP client, URL/IP parsers, Hono, LanceDB, Sharp, and related lockfile
+  resolutions; the production dependency audit is now clean.
+- **Large-store alias migration** -- startup project-ID normalization now uses a
+  bounded SQLite update instead of requiring the full observation corpus in
+  memory. The FTS5 triggers and storage generation stay current.
+- **Alias-aware search visibility** -- observations retained under a merged
+  project alias are readable through the same alias-expanded search scope.
+- **Hybrid search recovery** -- an empty or stale derived semantic result now
+  falls back to the durable lexical path instead of becoming a silent empty
+  answer.
+- **Search CLI parity** -- `memorix search "query"` and
+  `memorix search --query "query"` now share the same argument contract.
+- **Pi hooks on Node 26** -- hook stdin is paused and unrefed without closing
+  file descriptor 0, so subsequent synchronous Git detection remains safe when
+  a host supplies hook input through a regular file.
+
+### Added
+- **Qwen Code hooks** -- `memorix hooks install --agent qwen` writes the
+  official `.qwen/settings.json` hook shape, preserves user settings, and
+  installs project/global `QWEN.md` guidance. iFlow is not targeted because its
+  official CLI service was retired in April 2026.
+- **API Route memory preset** -- optional background memory LLM processing
+  through an OpenAI-compatible endpoint, with an isolated `API_ROUTE_API_KEY`,
+  CLI/TUI setup and bilingual configuration guidance.
+
+## [1.9.10] - 2026-10-08
+
+### Fixed
+- **OpenCode 1.x regression** -- generated plugins now include the v1
+  `server` field alongside the v2 `setup` contract, so OpenCode 1.x can load
+  the legacy named export. Lifecycle, file, command, message, and compaction
+  events are also forwarded through the v1 generic `event` hook, with a longer
+  Windows hook timeout to reduce stale Bun timer failures.
+
 ## [1.9.9] - 2026-10-02
 
 ### Added

package/README.md

@@ -535,6 +535,8 @@ For OpenRouter embeddings, use `provider = "api"`, `base_url = "https://openrout
 
 For Requesty, choose Requesty in `memorix init` or set `provider = "requesty"` in `[memory.llm]` (endpoint `https://router.requesty.ai/v1`, key from `REQUESTY_API_KEY` or `MEMORIX_LLM_API_KEY`). Requesty embeddings use `provider = "api"`, `base_url = "https://router.requesty.ai/v1"`, and `model = "openai/text-embedding-3-small"` with `REQUESTY_API_KEY`.
 
+For [API Route](https://www.api-route.com), choose API Route in `memorix init` or set `provider = "apiroute"` in `[memory.llm]`. The preset uses `https://global.api-route.com/v1` and `deepseek-v4.1-flash`, with your own `API_ROUTE_API_KEY` or `MEMORIX_LLM_API_KEY`. It is opt-in memory LLM processing; embeddings and memcode's model registry are configured separately. See [configuration](docs/CONFIGURATION.md#api-route-memory-preset).
+
 For controlled MiniMax media generation, set `MINIMAX_API_KEY` (global) or
 `MINIMAX_CN_API_KEY` (China region) in your environment or `.env`. The media
 library never stores that key, a signed output URL, or base64 payloads. CLI

package/README.zh-CN.md

@@ -508,6 +508,8 @@ OpenAI-compatible 的 `https://api.atlascloud.ai/v1` endpoint，并读取
 
 使用 Requesty 时，可在 `memorix init` 中选择 Requesty，或在 `[memory.llm]` 中设置 `provider = "requesty"`（endpoint 为 `https://router.requesty.ai/v1`，key 读取 `REQUESTY_API_KEY` 或 `MEMORIX_LLM_API_KEY`）。Requesty embedding 使用 `provider = "api"`、`base_url = "https://router.requesty.ai/v1"`、`model = "openai/text-embedding-3-small"`，并读取 `REQUESTY_API_KEY`。
 
+使用 [API Route](https://www.api-route.com) 时，可在 `memorix init` 中选择 API Route，或在 `[memory.llm]` 中设置 `provider = "apiroute"`。预设使用 `https://global.api-route.com/v1` 和 `deepseek-v4.1-flash`，密钥读取自己的 `API_ROUTE_API_KEY` 或 `MEMORIX_LLM_API_KEY`。这是显式启用的记忆 LLM 处理；embedding 和 memcode 模型注册表仍需单独配置。详见[配置说明](docs/CONFIGURATION.md#api-route-memory-preset)。
+
 受控 MiniMax 媒体生成可在环境变量或 `.env` 中设置全局 `MINIMAX_API_KEY`，或中国区 `MINIMAX_CN_API_KEY`。媒体库不会保存该 key、临时签名 URL 或 base64 负载。CLI 生成本身就是显式操作；若要让 MCP 中的 Agent 发起生成，还必须明确设置 `MEMORIX_MCP_MEDIA_GENERATION=1`。
 
 <h2 id="docker"><picture><source media="(prefers-color-scheme: dark)" srcset="assets/tags/light/section-docker.svg"><img src="assets/tags/section-docker.svg" alt="Docker" height="32" /></picture></h2>