Full Changelog: v0.4.19...v0.4.20

Fix cross-process model result reuse in the workflow CLI. The SDK's modality sets could serialize in different orders and generate different cache keys for identical requests. Cache keys now use canonical modality order and retain reads of existing order-sensitive entries.

Schema repair messages also include nested failure locations without logging rejected input values.

All three release platforms passed the backend checks, neutral report example, packaging and packaged CLI smoke checks. Includes the headless CLI introduced in v0.4.18.

CLI guide

Implementation: #114.

VectorVein now includes a headless workflow CLI and a console-mode VectorVeinCLI executable in the Windows, macOS and Linux packages.

Commands cover workspace initialization, workflow validation/creation/export/update/patching, synchronous execution, run status and records. The CLI shares the desktop data model and execution engine, emits UTF-8 JSON, supports isolated workspaces and external model settings, and preserves workflow versions in run records.

Model nodes support explicit output/concurrency limits, fixed-endpoint selection and schema-failure diagnostics. Each request reserves the rate limit once, and cleanup timers no longer delay CLI exit.

Validation: 60 backend tests passed on all three release platforms. Each packaged CLI passed help and workspace-initialization smoke checks. The existing neutral end-to-end report example also passed.

CLI commands and configuration

Implementation: #112. Release version: #113. The existing v0.4.17 tag remains unchanged.

VectorVein workflows now preserve multi-sheet spreadsheet data, reject incomplete model batches, and generate validated report artifacts.

• Read all or selected Excel worksheets as text or structured JSON with source paths and row numbers; export named, typed worksheets with explicit column formats.
• Validate LLM outputs with JSON Schema, perform bounded repairs, and optionally reuse successful results per item. Failed batches stop downstream execution.
• Propagate Python errors, reject mismatched batch lengths, execute top-level code once, and preserve subworkflow retry state.
• Export structured DOCX reports and headless PNG charts using atomic file writes.
• Add Chinese/English editor controls and a neutral reproducible workflow example.
• Update frontend dependencies and security pins; the dependency audit passes.

Compatibility: Python nodes now stop on errors by default. Workflows that intentionally route error_msg should explicitly enable continue_on_error. Spreadsheet string cells remain literal data.

Validation includes 57 backend tests, lint, frontend audit/build, and a neutral end-to-end DAG that exercises schema repair, source-total verification, report exports and a second run with no new model request. The example uses a loopback-only deterministic model stub and does not evaluate model quality.

Workflow guide and reproduction steps

Implementation: #109. Release lint policy: #111.

模型目录更新

• 升级至 vv-llm 0.7.4，从共享目录加载 16 家提供商、169 个可选聊天模型，新增小米 MiMo。
• 更新聊天、工作流和视觉节点的模型选项；视觉节点按原生图像输入能力筛选。
• 移除旧内置聊天模型与过期别名，更新默认模型。
• 自动迁移已有设置，保留自定义模型、API 端点绑定与启用状态。
• 适配新版 OpenAI / Anthropic SDK 与 httpx2，修正 MiniMax 国内默认 API 地址。
• 保留语音识别、语音合成及图像生成节点所需的服务路由。

升级提示

旧内置聊天模型已从可选列表退役。使用旧模型的已有工作流，请在节点中重新选择当前模型。

验证

• 后端测试：46 项通过。
• 前端改动文件的 ESLint、生产构建与模型目录集成检查通过。
• Windows、macOS、Linux 三平台 ZIP 构建及上传通过；Windows 包的版本、模型目录和 SDK 模块检查通过。
• OpenAI、Anthropic、DeepSeek、Qwen、智谱、Gemini、xAI、Moonshot、MiniMax 的真实流式 API 调用通过。

已知检查结果

独立的 Frontend quality 工作流在依赖审计步骤失败，现有锁定依赖仍有 31 条漏洞记录（12 高、17 中、2 低）。

完整变更：v0.4.14...v0.4.15

What's Changed

• chore(deps): bump smplr from 0.20.0 to 1.0.0 in /frontend by @dependabot[bot] in #80

Full Changelog: v0.4.13...v0.4.14

Full Changelog: v0.4.12...v0.4.13

Release Notes - Version 0.4.11

This release supersedes v0.4.10. Please use v0.4.11 for all desktop downloads.

Desktop Release Packaging:

• Fixed the Windows desktop package startup failure caused by missing packaged runtime dependencies such as numpy.
• Fixed the workflow editor node palette in packaged desktop builds. The left sidebar now correctly shows the workflow nodes instead of displaying "No matching nodes found" / "没有匹配的节点".
• Added release build safeguards so GitHub Actions fails immediately if the packaged workflow editor bundle is missing node registrations.
• Added packaging checks for backend runtime dependencies, PyInstaller hidden imports, lockfile groups, and frontend workflow editor assets to reduce the chance of publishing unusable desktop packages.
• The Windows, macOS, and Linux desktop packages for this release were built and published by GitHub Actions.

Workflow Branching:

• The Conditional node supports ordered multi-branch matching, so a single node can route different inputs into different downstream paths.
• Only the matched branch continues execution. Unmatched branches and their downstream nodes are marked as skipped instead of continuing to run.
• Added backend branch resolution and regression tests to cover branch matching, skip propagation, and downstream execution control.

Run Status & Debugging:

• Fixed workflow use pages so outputs on skipped branches no longer remain stuck in loading state after the matched branch has already finished.
• Fixed workflow completion settlement after branch execution so the run button no longer stays in a running state when the workflow has already ended.
• Debug and run-record views now show skipped nodes as skipped instead of rendering a green success header and misleading execution time.
• Fixed fast branch-run status polling so check-status no longer gets stuck on stale 202 responses in this scenario.

Model Defaults:

• Updated the default MiniMax API base URL.

工作流分支:

• Conditional 条件判断节点现已支持按顺序匹配多个分支，一个节点即可根据不同输入分流到不同下游路径。
• 只有命中的分支会继续执行，未命中的分支及其下游节点现在会被明确标记为“已跳过”，不再继续运行。
• 补充了后端分支解析与回归测试，覆盖分支命中、跳过传播和下游执行控制等场景。

运行状态与调试:

• 修复工作流使用页中被跳过分支的输出节点仍然卡在 loading 的问题，命中分支完成后页面状态可以正常收口。
• 修复分支执行结束后运行按钮仍停留在“运行中”的问题，工作流完成状态现在可以正确结束。
• 调试视图和运行记录中的被跳过节点现在会显示为“已跳过”，不再错误显示绿色成功头和误导性的运行时间。
• 修复快速分支运行场景下 check-status 轮询可能卡在过期 202 响应的问题。

桌面端发布包:

• 修复 Windows 桌面发布包启动时报缺少 numpy 等运行时依赖的问题。
• 修复打包后的桌面端工作流编辑器左侧节点面板为空、显示“没有匹配的节点”的问题。
• 新增发布构建守卫：如果 GitHub Actions 构建出的工作流编辑器前端 bundle 缺少节点注册，会立即失败，不再发布坏包。
• 新增后端运行时依赖、PyInstaller hidden import、锁文件分组和前端节点注册等发布校验，降低发布不可用桌面包的风险。
• 本版本 Windows、macOS、Linux 桌面包均由 GitHub Actions 自动构建并发布。

模型默认配置:

• 更新 MiniMax 默认 API 地址。

Full Changelog: v0.4.6...v0.4.11

⚠️ Superseded Release - Do Not Use

This release has been superseded by v0.4.11.

Known issue:

• The packaged desktop workflow editor can show an empty left node palette with "No matching nodes found" / "没有匹配的节点" because the production frontend bundle was missing workflow node registrations.

Please download v0.4.11 or newer instead:
https://github.com/AndersonBY/vector-vein/releases/tag/v0.4.11

Full Changelog: v0.4.9...v0.4.10

Release Notes - Version 0.4.6

Workflow Branching:

• The Conditional node now supports ordered multi-branch matching, so a single node can route different inputs into different downstream paths.
• Only the matched branch continues execution. Unmatched branches and their downstream nodes are now marked as skipped instead of continuing to run.
• Added backend branch resolution and regression tests to cover branch matching, skip propagation, and downstream execution control.

Run Status & Debugging:

• Fixed workflow use pages so outputs on skipped branches no longer remain stuck in loading state after the matched branch has already finished.
• Fixed workflow completion settlement after branch execution so the run button no longer stays in a running state when the workflow has already ended.
• Debug and run-record views now show skipped nodes as skipped instead of rendering a green success header and misleading execution time.
• Fixed fast branch-run status polling so check-status no longer gets stuck on stale 202 responses in this scenario.

工作流分支:

• Conditional 条件判断节点现已支持按顺序匹配多个分支，一个节点即可根据不同输入分流到不同下游路径。
• 只有命中的分支会继续执行，未命中的分支及其下游节点现在会被明确标记为“已跳过”，不再继续运行。
• 补充了后端分支解析与回归测试，覆盖分支命中、跳过传播和下游执行控制等场景。

运行状态与调试:

• 修复工作流使用页中被跳过分支的输出节点仍然卡在 loading 的问题，命中分支完成后页面状态可以正常收口。
• 修复分支执行结束后运行按钮仍停留在“运行中”的问题，工作流完成状态现在可以正确结束。
• 调试视图和运行记录中的被跳过节点现在会显示为“已跳过”，不再错误显示绿色成功头和误导性的运行时间。
• 修复快速分支运行场景下 check-status 轮询可能卡在过期 202 响应的问题。

Full Changelog: v0.4.5...v0.4.6