GLM问问 v0.1.17

Fixes

• Parse Claude Code's object-shaped modelUsage metadata so the context-window progress bar appears after Agent replies.
• Lower the ask panel from screen-saver to floating, while keeping the transient selection toolbar above fullscreen windows.

Tests

• Added Claude context metadata regression coverage.
• Added macOS window-level regression coverage.

GLM问问 v0.1.16

Features

• Added the current context-window usage and progress bar to the token usage popover.
• Context usage uses exact server-reported token data; Claude Code context size comes from its contextWindow metadata, and unsupported agents hide the section instead of estimating.

Tests

• Added context usage formatting and persistence regression checks.
• Verified the full build and release packaging pipeline.

GLM问问 v0.1.15

Fixes

• Speed up intent classification by trying disabled thinking first, with automatic low-thinking fallback only when the endpoint requires it.
• Retry intent classification with low thinking when a fast-path response returns HTTP 200 but invalid JSON, and count both real usages.
• Preserve unfinished replies as retryable interruptions after an unexpected app exit.
• Normalize interrupted sessions in both main-process history reads and renderer session restoration.
• Keep the original topic running when the user creates a new topic immediately after sending a message.

Tests

• Added intent classification adaptive-thinking and invalid-JSON retry checks.
• Added interrupted-session restoration checks.
• Added background stream ownership regression checks.

GLM问问 v0.1.14

Fixes

• Clear the in-popup copy/follow-up selection bubble when a text selection opens a fresh conversation.
• Hide the in-popup selection bubble when the input gains focus or the user starts typing.
• Hide the external selection toolbar when the user starts typing in any application.
• Cancel pending keyboard-selection probes when typing begins, preventing a delayed toolbar from reappearing.

Tests

• Added selection-bubble regression checks for fresh sessions, typing, focus behavior, and pending selection cancellation.

GLM问问 v0.1.13

Features

• Show exact server-reported token usage for each reply, including prompt and completion totals.
• Include intent-classification requests in per-reply token totals.
• Parse exact usage reported by Claude Code and Codex local Agent runs.
• Add a persistent per-session Tokens badge with exact cumulative consumption and an details popover.
• Keep cumulative usage persisted across session reloads; edits, retries, and deletions do not deduct actual consumption.
• Reset the session token badge when a text selection starts a fresh conversation.

Fixes

• Use a low-reasoning intent-classification request for models that cannot disable thinking.
• Keep the token popover at a fixed width and center it above the Tokens button.

Tests

• Added token usage propagation, aggregation, UI, positioning, and fresh-session regression checks.

GLM问问 v0.1.12

Fixes

• Remove the 4-minute total execution limit for local Agents; long-running Agents can continue while output remains active.
• Keep the existing 60-second inactivity guard for stalled Agent connections.
• Restore partial mouse text selection in the main input box while preserving keyboard selection and editing behavior.

Tests

• Added regression checks for Agent timeout behavior and input selection.

GLM问问 v0.1.11

Fixes

• Retry and regenerate now re-enter intent routing instead of forcing GLM direct response.
• Restore each topic's scroll position when switching between sessions.
• Improve nested tool-trace wheel scrolling and duplicate session-focus handling.
• Tool trace status now reflects the final reply state instead of individual tool failures.

Tests

• Added regression checks for retry routing, scroll behavior, and tool status.

GLM问问 v0.1.10

修复与改进

• 图片请求不再向 GLM-5.3 发送 image_url，会保留/落盘本地图片路径，并交给可用本地 Agent 完成识别、解答与回复。
• 粘贴图片、选择图片、编辑重发、失败重试和重新生成均保留本地 Agent 视觉链路。
• 双击/三击链路按最后一次点击捕获，避免连续多击时工具条重复弹出。
• 多击捕获提前并行执行，并将展示时间统一到约 500ms，减少 AX 应用与微信等 Copy 兜底应用的体感差异。

安装

下载 GLMAsk-0.1.10-arm64.dmg 或 GLMAsk-0.1.10-arm64-mac.zip。应用未签名，首次打开请右键选择「打开」。

GLM问问 v0.1.9

修复与改进

• 划词识别重做为三级策略：AX / Text Marker 直读 → 目标 App 编辑菜单 Copy → 目标进程 Cmd+C，只要最终识别到真实字符串就会展示工具条。
• 支持键盘选词：Cmd+A、Shift+方向键/Home/End/PageUp/PageDown。
• 降低鼠标拖选阈值，短选区不再漏触发。
• 修复微信等 AX 不暴露选区 App 的明显延迟：跳过重复 AX 探测，尽快进入 Copy 兜底。
• 复制类兜底会完整备份/恢复剪贴板 flavors，并继续排除对象拖拽、Finder、截图与禁用 Copy 菜单场景。
• 修复工具条点击后误激活 Finder、问一问弹窗切换桌面等问题。
• 历史记录默认分页加载，最多保留 1000 条。

安装

下载 GLMAsk-0.1.9-arm64.dmg 或 GLMAsk-0.1.9-arm64-mac.zip，将 GLM问问.app 拖到应用程序目录后打开。

应用未签名，首次打开请右键 → 打开，并在系统设置中授予辅助功能权限。

GLM问问 v0.1.8

本版修复：

• 编辑文案后立即显示“正在分析请求意图…”，消除意图识别期间数秒无响应的空窗
• 编辑重发改为单一请求生命周期：点击编辑不打断当前回复，点击发送才替换旧 turn
• 修复快捷键唤起时 macOS panel 不激活所属应用，导致访达仍保持活跃的问题

下载 GLMAsk-0.1.8-arm64.dmg 或 GLMAsk-0.1.8-arm64-mac.zip。应用未签名，首次打开请右键选择「打开」。