feat(multimodal): support document image extraction and inline vision processing

Added document image extraction capability for PDF and DOCX files, including page/index metadata and storage integration. Extended `process_files` with `document_image_recognition` flag to conditionally enable vision-based image processing when model supports it. Updated knowledge repository and workflow node logic to enforce status=1 checks. Added PyMuPDF dependency.
This commit is contained in:
Timebomb2018
2026-04-24 11:18:50 +08:00
parent 9fdb952396
commit 767eb5e6f2
13 changed files with 397 additions and 52 deletions

View File

@@ -1,4 +1,36 @@
{
"v0.3.1": {
"introduction": {
"codeName": "无境",
"releaseDate": "2026-4-22",
"upgradePosition": "🐻 聚焦应用体验优化、记忆 API 开放与工作流可靠性提升,打破边界,自由流动",
"coreUpgrades": [
"1. 应用与模型增强<br>* 模型 Key 全删后自动关闭:避免无 Key 运行时错误<br>* 模型 JSON 格式化输出开关:支持旧工作流迁移的稳定 JSON 输出<br>* 配置导入覆盖:支持完整替换当前配置<br>* 导入时缺失资源清理:自动清空不存在的工具和知识库引用",
"2. 记忆 API 与智能 📚<br>* 记忆读写 API 与 End-User Key 供给:支持第三方直接交互记忆层<br>* 记忆库 API 与配置更新:程序化控制记忆设置(提供顺序接口)<br>* End-User 元数据存储:丰富用户上下文持久化",
"3. 工作流与体验优化 ⚙️<br>* 会话历史文件元数据:增加文件大小、名称和类型<br>* 迭代节点并行输入修复:恢复并发执行行为<br>* API Key 后四位展示:便于密钥识别<br>* 条件分支多文件子变量:更精细的条件逻辑<br>* Agent 模型配置重置接口:完善前后端契约<br>* 三级变量键盘导航:提升变量选择体验<br>* 应用标签页动态标题:动态显示应用名称<br>* 变量聚合三级勾选修复:修复勾选行为<br>* 工作流检查清单校验增强:工具必填和视觉变量必填<br>* 变量聚合器到参数提取器输出:修复输出变量获取",
"4. 知识库与性能 ⚡<br>* 文档解析与 Graph 异步执行:提升文档摄入吞吐量",
"5. 稳健性与缺陷修复 🔧<br>* 工具节点原始参数类型:修复类型不匹配问题<br>* 前端部署后资源过期导入错误:解决缓存资源导入失败<br>* 工作流工具节点必填校验:防止不完整配置发布",
"<br>",
"v0.3.1 是平台哲学演进中的关键时刻——边界的打破。记忆 API 开放和应用体验优化为社区用户提供更强大的集成能力。展望未来,我们将持续提升记忆智能管线的萃取精度与自适应遗忘策略,深化工作流引擎能力。破界而行,臻于无境。",
"MemoryBear — 无境 🐻✨"
]
},
"introduction_en": {
"codeName": "WuJing",
"releaseDate": "2026-4-22",
"upgradePosition": "🐻 Focused application improvements, memory API openness, and workflow reliability — dissolving boundaries, flowing freely",
"coreUpgrades": [
"1. Application & Model Enhancements<br>* Model Auto-Disable on Key Deletion: Prevents keyless runtime errors<br>* Model JSON Formatted Output Toggle: Stable JSON output for legacy workflow migration<br>* Configuration Import with Override: Full configuration replacement support<br>* Import Cleanup for Missing Resources: Auto-clears missing tool and knowledge base references",
"2. Memory API & Intelligence 📚<br>* Memory Read/Write API with End-User Key Provisioning: Third-party memory layer interaction<br>* Memory Store API & Configuration Update: Programmatic memory settings control with sequential interface<br>* End-User Metadata Storage: Richer user context persistence",
"3. Workflow & UX Improvements ⚙️<br>* Conversation History File Metadata: File size, name, and type labels<br>* Iteration Node Parallel Input Fix: Restored concurrent execution<br>* API Key Last Four Digits Display: Key identification without exposure<br>* Condition Branch Multi-File Sub-Variables: Granular conditional logic<br>* Agent Model Config Reset Endpoint: Completed frontend-backend contract<br>* Three-Level Variable Keyboard Navigation: Improved selection experience<br>* Dynamic Tab Title for Applications: Dynamic app name in browser tab<br>* Variable Aggregator Three-Level Checkbox Fix: Corrected checkbox behavior<br>* Workflow Checklist Validation Enhancements: Tool required and vision variable validation<br>* Variable Aggregator to Parameter Extractor Output: Fixed output variable access",
"4. Knowledge Base & Performance ⚡<br>* Async Document Parsing & Graph Execution: Improved document ingestion throughput",
"5. Robustness & Bug Fixes 🔧<br>* Tool Node Raw Parameter Types: Fixed type mismatch issues<br>* Stale Frontend Resource Import Error: Resolved cached resource import failure<br>* Workflow Tool Node Required Validation: Prevents incomplete configuration publishing",
"<br>",
"v0.3.1 marks a pivotal moment in the platform's evolution — the dissolution of boundaries. Memory API openness and application experience improvements provide community users with stronger integration capabilities. Looking ahead, we will continue improving extraction accuracy, adaptive forgetting strategies, and deepening workflow engine capabilities. Beyond boundaries — the boundless awaits.",
"MemoryBear — The Boundless 🐻✨"
]
}
},
"v0.3.0": {
"introduction": {
"codeName": "破晓",