wisp-science v1.8.0
v1.8.0 turns persistent Python and R runtimes into an integrated analysis workbench. Edit and run project scripts, inspect live objects, use an interactive console, capture plots, and execute reproducible saved scripts against large objects that are already loaded in memory.
Downloads
- macOS (.dmg, signed + notarized): choose the Apple Silicon (aarch64) or Intel (x86_64) build; compatible versions can update in-app.
- Windows installers (MSI / NSIS): published unsigned while SignPath release signing is disabled; Windows may show SmartScreen, so review the publisher details before choosing "Run anyway".
- Linux (deb / AppImage): available for both x86_64 and aarch64.
See Assets below.
Highlights
Python & R Runtime Workbench
- Open an RStudio-style workbench with editable
.pyand.Rsource, a multiline console with history, a live variable inspector, a plots pane, and draggable quadrant dividers. - Run a selected region or the complete statement at the caret with Ctrl/Command+Enter. Multiline R calls, pipes, control flow, and Python suites are kept together, while the editor advances to the next statement.
- Run the complete saved script from the editor with Ctrl/Command+Shift+Enter. Dirty drafts are saved first, and execution stays inside the bound persistent Runtime instead of spawning a fresh
pythonorRscriptprocess. - Python matplotlib figures and R display-list plots are captured after execution and can be paged or cleared without taking over explicit file devices such as
png()orggsave(). - Conversation runtime chips show local and attached/default remote environments, update through Busy and Ready states while the agent works, and open the live Python or R object table directly.
- Choose the global default analysis environment from Settings or the Compute menu, with attached and auto-attached servers labeled separately.
Reproducible Scripts on Persistent State
- Agent Python/R tools can execute a project-local
script_pathinside an already-running Runtime, preserving large in-memory objects instead of reloading them in a fresh process. required_objectsguards fail before evaluation when a dependency is missing, andexpected_runtime_generationrejects a restarted or replaced Runtime.- Script results report the project path, SHA-256 hash, Runtime identity, and generation; Python and R diagnostics retain the source filename.
- Bound
.pyand.Rlinks now create immutable, syntax-highlighted snapshots. Chat-opened snapshots stay immutable and can switch to the live workspace editor when the original project source is available.
Workflow & File Improvements
- Plans and approval steps render full Markdown, while supported OpenAI-compatible conversations gain an optional per-conversation Fast mode using the provider's priority service tier.
- Model profiles can keep different API keys for different models on the same Base URL without overwriting one another.
- The Files panel can sort local and remote listings by name, size, or modification time.
- Right-click a recognized project path in chat to open it in the center, copy its absolute or relative path, or reveal it in Finder/Explorer.
Reliability
- Runtime status and open memory inspectors refresh after agent Python/R cells without losing filters or last-known object snapshots.
- Pinned runtime inspectors stay on screen after the window is resized.
- Source selections remain visible while dragging, UTF-16 offsets stay aligned for CJK text and emoji, and the console supports multiline input without losing its history.
- Selection popups, quote actions, artifact-version downloads, and script snapshot layouts remain stable across runtime panels, follow-scroll, branches, and explorations.
Notes
- Runtime console output is buffered per execution rather than streamed.
- R captures the current plot page for each cell; Python closes harvested matplotlib figures after the cell. Plot capture degrades gracefully when a headless R environment has no PNG-capable device.
- Whole-script execution from the editor is available for local project
.pyand.Rfiles. Remote and immutable artifact tabs remain read-only execution sources.
Full Changes
All commits since v1.7.1:
v1.7.1...v1.8.0
wisp-science v1.8.0
v1.8.0 将持久 Python 与 R Runtime 升级为集成式分析工作台。你可以编辑并运行项目脚本、检查实时对象、使用交互式控制台、捕获绘图,并让可复现脚本直接复用内存中已经加载的大型对象。
下载
- macOS(.dmg,已签名 + 公证):请选择 Apple Silicon(aarch64)或 Intel(x86_64)对应架构;兼容版本可从应用内更新。
- Windows 安装包(MSI / NSIS):当前在 SignPath 正式签名停用期间以未签名形式发布;Windows 可能显示 SmartScreen,请核对发布者信息后再选择「仍要运行」。
- Linux(deb / AppImage):提供 x86_64 与 aarch64 两个架构。
见下方 Assets。
重点功能
Python 与 R Runtime 工作台
- 打开类似 RStudio 的工作台:可编辑
.py/.R源码、多行控制台与历史记录、实时变量检查器、绘图面板,以及可拖动的四象限分隔条。 - 使用 Ctrl/Command+Enter 运行选区或光标所在的完整语句。多行 R 调用、管道、控制流和 Python 代码块会作为整体执行,随后光标前进到下一条语句。
- 使用 Ctrl/Command+Shift+Enter 从编辑器运行整个已保存脚本。未保存草稿会先落盘,脚本在已绑定的持久 Runtime 内执行,不会另起新的
python或Rscript进程。 - Python matplotlib 图形与 R display-list 绘图会在执行后自动捕获,可翻页或清空,同时不会接管
png()、ggsave()等显式文件设备。 - 对话中的 Runtime 状态条会显示本地及已附加/默认的远程环境,在智能体运行时实时更新 Busy/Ready 状态,并可直接打开 Python 或 R 对象表。
- 可从设置或 Compute 菜单选择全局默认分析环境,并清楚区分已附加与自动附加的服务器。
在持久状态上执行可复现脚本
- 智能体的 Python/R 工具可在已运行的 Runtime 中执行项目内
script_path,复用大型内存对象,无需在新进程中重复加载。 required_objects会在依赖对象缺失时于求值前失败;expected_runtime_generation会拒绝已经重启或被替换的 Runtime。- 脚本结果会报告项目路径、SHA-256、Runtime 标识与代次;Python/R 诊断会保留源文件名。
.py与.R链接现在可绑定为不可变、带语法高亮的快照。聊天中打开的仍是不可变快照;若原始项目源码可用,可切换到实时工作区编辑器。
工作流与文件改进
- 计划与审批步骤支持完整 Markdown;受支持的 OpenAI 兼容对话可按会话开启 Fast 模式,使用供应商的 priority service tier。
- 同一 Base URL 下的不同模型可保存不同 API key,互不覆盖。
- Files 面板可按名称、大小或修改时间排序本地与远程目录。
- 在聊天中右键已识别的项目路径,可在中央区域打开、复制绝对/相对路径,或在 Finder/资源管理器中显示。
可靠性
- 智能体执行 Python/R 单元后,Runtime 状态与已打开的内存检查器会自动刷新,同时保留过滤条件和最近一次对象快照。
- 调整窗口大小后,已固定的 Runtime 检查器仍会保持在屏幕内。
- 拖动源码选区时高亮即时可见,中文与 emoji 的 UTF-16 偏移保持对齐;控制台支持多行输入且不会丢失历史。
- Runtime 面板、跟随滚动、分支与 Exploration 场景下的选区弹窗、引用操作、artifact version 下载和脚本快照布局更加稳定。
说明
- Runtime 控制台仍按每次执行缓冲输出,暂不流式展示。
- R 每个单元捕获当前绘图页;Python 在单元结束后关闭已收集的 matplotlib 图形。无 PNG 能力的无头 R 环境会跳过绘图捕获,但不影响代码执行。
- 编辑器整脚本执行适用于本地项目中的
.py与.R文件;远程与不可变 artifact 标签仍保持只读执行来源。
完整变更
自 v1.7.1 以来的全部提交见:
v1.7.1...v1.8.0