Releases: torlyai/desktop
TorlyAI Desktop v2.6.4
TorlyAI Desktop v2.6.3
TorlyAI Desktop v2.6.3
Release Date: 2026-03-20
Commits: 44 | Files Changed: 103 | +8,146 / -1,588 lines
Highlights
- ChatGPT OAuth Token Auto-Refresh — Fixed "Provided authentication token is expired" error that affected all ChatGPT Plus users after ~1 hour. Tokens now refresh automatically with a 5-minute preemptive buffer.
- Claude OAuth 429 Fix — Cherry-picked from upstream v0.7.10. Replaced spoofed browser User-Agent with proper client identifier, preventing rate limiting on Claude Max sign-in.
- 1M Context Window — Opus 4.6 and Sonnet 4.6 now use the full 1,000,000 token context window (up from 200K). Activated via
[1m]model suffix at the API level. - Spawn ENOENT Auto-Recovery — Sessions now automatically recover when the SDK binary moves during auto-update, instead of failing silently.
- 7-Day Autopilot Trial — Premium subscribers now receive a complimentary 7-day Autopilot trial upon subscription activation.
New Features
1M Context Window for Claude Models
feat(agent): Register Opus 4.6 and Sonnet 4.6 with 1,000,000 token context window in model registryfeat(agent): Append[1m]model suffix to API requests for models supporting 1M context — works for both API key and OAuth auth paths- Cherry-picked from upstream v0.7.8 (model registry) and v0.7.9 (API activation)
Autopilot Trial for Premium Users
feat(auth): Add 7-day Autopilot trial automatically activated for new Premium subscribers- Trial plan overlays the base Premium plan in JWT license claims (
trial_plan: 'autopilot',trial_ends_at)
Prospect Management (moved to torlyAI)
feat(prospects): Added prospect database, scrapers, and web UI (subsequently moved to torlyAI website repo)chore: Removed prospects from desktop repo — campaigns and admin tools belong in the website repo
Bug Fixes
Authentication & OAuth
fix(auth): ChatGPT OAuth token refresh — OpenAIAgent now callsrefreshChatGptTokens()when the access token expires, mirroring the logic from the deprecated PiAgent. Includes mutex to prevent concurrent refreshes, supports all 3 providers (ChatGPT, Copilot, Antigravity).fix(auth): Claude OAuth 429 fix — Cherry-picked from upstream v0.7.10. Replaced spoofed browser User-Agent header (Mozilla/5.0...) with proper client identifier (TorlyAI/{version}). Updated TOKEN_URL fromconsole.anthropic.comtoplatform.claude.com.
Agent System
fix(agent): Spawn ENOENT auto-retry — Detects when the SDK subprocess binary has been moved/deleted (e.g., during auto-update bundle swap) and automatically retries after 2 seconds with fresh paths.fix(agent): Synthetic model fallback for Pi providers with unknown model variantsfix(agent): Align OpenAI file write path with session directory and detect BP sectionsfix(agent): Remove project-outputs capability that caused files to save to wrong directoryfix(agent): Always save deliverables to session outputs path
UI
fix(ui): URL linkify fix — Cherry-picked from upstream v0.7.5. Fixes broken link rendering when URLs are wrapped in markdown bold/italic formatting.fix(ui): Scroll session list to top when returning from Project view
Project Workspace
fix(project): Bulletproof AI task parsing for ChatGPT/Pi providers — handles chatty LLM responses, BOM characters, and wrapper textfix(skills): Load AskUserQuestion deferred tool before interactive stepsfix(skills): Add ToolSearch to allowed-tools so AskUserQuestion can load
Build & Distribution
fix(build): beforeBuild OOM fix, arch-specific bun dirs, and build-dmg improvementsfix(build): Bundle arch-specific Bun binary per macOS DMG (prevents EBADARCH -86 on Intel Macs)fix(build): Copy all interceptor dependencies for asar unpacking
Performance Improvements
perf(living-bp): Add caching layer and reduce IPC payload for Living Business Plan updatesperf(living-bp): Debounce living-bp-updated SSE/IPC events to reduce UI re-renders
Infrastructure
refactor(project): Renameproject/outputs/→project/exports/to resolve naming collision with agent output directoryrefactor(release): Consolidate release commands into 3-command workflow (/release-macos-2-pack,/release-windows-3-pack,/release-notes)- Cherry-picked 4 upstream fixes from lukilabs/craft-agents-oss (v0.7.1, v0.7.5, v0.7.8/v0.7.9, v0.7.10)
Documentation
- VIP50 + assessment lifecycle emails implementation plan and design spec
- Apple Developer cookbook updates
- GitHub integration architecture review
- Build & distribution cookbook updates
Download
| Platform | Download | Requirements |
|---|---|---|
| macOS (Apple Silicon) | TorlyAI-2.6.3-arm64.dmg | macOS 13+ (M1/M2/M3/M4) |
| macOS (Intel) | TorlyAI-2.6.3-x64.dmg | macOS 13+ (Intel) |
| Windows (x64) | TorlyAI-Setup-2.6.3-x64.exe | Windows 10+ (Intel/AMD) |
| Windows (ARM64) | TorlyAI-Setup-2.6.3-arm64.exe | Windows 11 ARM |
TorlyAI Desktop v2.6.3 — 发布说明
发布日期: 2026年3月20日
提交数: 44 | 变更文件: 103 | +8,146 / -1,588 行
亮点
- ChatGPT OAuth 令牌自动刷新 — 修复了所有 ChatGPT Plus 用户在使用约1小时后遇到的"身份验证令牌已过期"错误。现在令牌会在过期前5分钟自动刷新,用户无需手动重新登录。
- Claude OAuth 429 修复 — 从上游 v0.7.10 移植。将伪造的浏览器 User-Agent 替换为正确的客户端标识符,防止 Claude Max 登录时触发速率限制。
- 100万上下文窗口 — Opus 4.6 和 Sonnet 4.6 现在使用完整的 1,000,000 令牌上下文窗口(从20万提升)。通过 API 层面的
[1m]模型后缀激活。 - 进程异常自动恢复 — 当自动更新替换应用程序文件时,会话现在能够自动恢复,而不是静默失败。
- 7天 Autopilot 试用 — Premium 订阅用户现在在激活订阅后自动获得7天的 Autopilot 试用体验。
新功能
Claude 模型100万上下文窗口
- 在模型注册表中将 Opus 4.6 和 Sonnet 4.6 的上下文窗口更新为 1,000,000 令牌
- 对支持100万上下文的模型,API 请求自动附加
[1m]后缀——适用于 API 密钥和 OAuth 两种认证方式 - 从上游 v0.7.8(模型注册表)和 v0.7.9(API 激活)移植
Premium 用户 Autopilot 试用
- 新 Premium 订阅用户自动激活7天 Autopilot 试用
- 试用计划通过 JWT 许可证声明覆盖基础 Premium 计划
问题修复
身份验证与 OAuth
- ChatGPT OAuth 令牌刷新 — OpenAIAgent 现在在访问令牌过期时自动调用
refreshChatGptTokens(),从 OpenAI 令牌端点获取新令牌。包含互斥锁防止并发刷新,支持 ChatGPT、Copilot 和 Antigravity 三种提供商。 - Claude OAuth 429 修复 — 从上游 v0.7.10 移植。将伪造的浏览器 User-Agent 替换为
TorlyAI/{版本号}客户端标识符。
AI 代理系统
- 进程 ENOENT 自动重试 — 检测 SDK 子进程二进制文件在自动更新期间被移动或删除的情况,2秒后自动使用新路径重试。
- 修复 Pi 提供商未知模型变体的合成模型回退
- 修复 OpenAI 文件写入路径与会话目录对齐
- 修复交付物始终保存到会话输出路径
用户界面
- URL 链接修复 — 从上游 v0.7.5 移植。修复了 URL 被 Markdown 粗体或斜体包裹时链接渲染异常的问题。
- 从项目视图返回时,会话列表自动滚动到顶部
项目工作区
- 增强 ChatGPT/Pi 提供商的 AI 任务解析鲁棒性——处理冗余 LLM 响应、BOM 字符和包装文本
构建与分发
- beforeBuild OOM 修复,架构特定 bun 目录
- 每个 macOS DMG 捆绑对应架构的 Bun 二进制文件(防止 Intel Mac 上的 EBADARCH -86 错误)
性能优化
- Living Business Plan 更新增加缓存层,减少 IPC 负载
- 防抖 Living BP 更新事件,减少 UI 重新渲染
下载
| 平台 | 下载链接 | 系统要求 |
|---|---|---|
| macOS (Apple Silicon) | TorlyAI-2.6.3-arm64.dmg | macOS 13+(M1/M2/M3/M4) |
| macOS (Intel) | TorlyAI-2.6.3-x64.dmg | macOS 13+(Intel) |
| Windows (x64) | TorlyAI-Setup-2.6.3-x64.exe | Windows 10+(Intel/AMD) |
| Windows (ARM64) | TorlyAI-Setup-2.6.3-arm64.exe | Windows 11 ARM |
TorlyAI Desktop v2.6.2
TorlyAI Desktop v2.6.2
Release Date: 8 March 2026 | Commits: 13 | 38 files changed (+2,474 / −148)
Highlights
- Build pipeline hardening — Added
beforeBuild.cjshook to prevent electron-builder OOM on large monorepos, switched to arch-specific Bun binary directories, and pinned electron-builder to v26.8.1. - Synthetic model fallback — Pi providers with unknown model variants now gracefully fall back instead of crashing sessions.
- Skill pipeline fix — Interactive skills (e.g. release commands) now properly load deferred tools like AskUserQuestion before prompting.
- AI task parsing resilience — ChatGPT/Pi providers no longer fail on malformed task parsing responses.
Bug Fixes
- fix(build):
beforeBuild.cjsOOM prevention hook, arch-specific Bun dirs (vendor/bun-arm64/,vendor/bun-x64/), and DMG naming improvements (1f5b444) - fix(agent): Synthetic model fallback for Pi providers with unknown model variants — prevents session crash on custom fine-tunes or new model strings (
8a0b900) - fix(ui): Scroll session list to top when returning from Project Workspace view (
a7db55b) - fix(project): Bulletproof AI task parsing for ChatGPT/Pi providers — handles malformed JSON and missing fields gracefully (
68be1b4) - fix(skills): Load AskUserQuestion deferred tool before interactive steps in skill pipeline (
0276274) - fix(skills): Add ToolSearch to allowed-tools list so deferred tools can be discovered during skill execution (
17ab4f9)
Infrastructure
- chore: Bump version to 2.6.2 (
3ae4dd6,d2db40f)
Documentation
- docs: Add release notes for v2.6.1 (
ac5cc7f) - docs: Update macOS download links to v2.6.1 in README (
3fffd1b) - docs: Update Windows download links to v2.6.1 in README (
a280918) - docs: Update download links to v2.6.2 in README (
a76f5e9) - docs: Add release notes for v2.6.2 (
a99853c)
Downloads
| Platform | Download |
|---|---|
| macOS (Apple Silicon) | TorlyAI-2.6.2-arm64.dmg |
| macOS (Intel) | TorlyAI-2.6.2-x64.dmg |
| Windows (x64) | TorlyAI-Setup-2.6.2-x64.exe |
| Windows (ARM64) | TorlyAI-Setup-2.6.2-arm64.exe |
Full changelog: v2.6.1...v2.6.2
TorlyAI Desktop v2.6.1
TorlyAI Desktop v2.6.1
Release Date: 6 March 2026 | Commits: 20 | 66 files changed (+3,666 / −1,389)
Highlights
-
Autopilot diagnostic messages — The Start Autopilot dialog now explains why no tasks are ready (missing team member, not in Working column, no tasks at all) instead of a generic "No tasks ready" message. Info (ⓘ) tooltips on key terms guide new users through the two prerequisites.
-
Living BP performance boost — Debounced SSE/IPC events and added a caching layer to reduce payload size, making the real-time business plan preview noticeably faster on large documents.
-
Sidebar panels & Portal navigation — New Portal Site and Patreon Community panels accessible from the sidebar, plus UI polish across project navigation.
-
Agent file-write path fix — Deliverables from all providers (Claude, ChatGPT/Codex, Copilot) now save correctly to the session outputs directory. Fixed an issue where OpenAI-backend files landed in the wrong folder.
-
Arch-specific Bun binaries — macOS DMGs now bundle the correct Bun binary per architecture (arm64 vs x64), preventing EBADARCH errors on Intel Macs.
-
Build interceptor fix — All network interceptor dependencies are now copied for ASAR unpacking, resolving a startup crash on fresh installs.
-
Project exports rename — Renamed
project/outputs/→project/exports/to resolve a naming collision with session outputs.
Downloads
| Platform | Download |
|---|---|
| macOS (Apple Silicon) | TorlyAI-2.6.1-arm64.dmg |
| macOS (Intel) | TorlyAI-2.6.1-x64.dmg |
| Windows (x64) | TorlyAI-Setup-2.6.1-x64.exe |
| Windows (ARM64) | TorlyAI-Setup-2.6.1-arm64.exe |
Full changelog: v2.6.0...v2.6.1
TorlyAI Desktop v2.6.1 — 发布说明
发布日期: 2026年3月6日 | 提交数: 20 | 66个文件变更 (+3,666 / −1,389)
更新亮点
-
自动驾驶诊断消息 — "启动自动驾驶"对话框现在会明确告知用户任务未就绪的原因(未分配团队成员、未移至"进行中"列、或没有任务),而不再显示笼统的"无可用任务"提示。关键术语旁增加了 ⓘ 信息图标,帮助新用户了解两项前置条件。
-
实时商业计划书性能优化 — 对 SSE/IPC 事件进行防抖处理,并新增缓存层以减小数据负载,在大型文档上实时预览明显更流畅。
-
侧边栏面板与门户导航 — 新增"门户网站"和"Patreon 社区"面板,可从侧边栏直接访问,同时优化了项目导航的整体视觉体验。
-
代理文件写入路径修复 — 所有后端(Claude、ChatGPT/Codex、Copilot)生成的交付物现在均正确保存至会话输出目录。修复了 OpenAI 后端文件存储位置错误的问题。
-
架构专属 Bun 二进制文件 — macOS DMG 现在按架构(arm64 与 x64)打包正确的 Bun 二进制文件,避免 Intel Mac 上出现 EBADARCH 错误。
-
构建拦截器修复 — 所有网络拦截器依赖项现已正确复制用于 ASAR 解包,解决了全新安装时的启动崩溃问题。
-
项目导出目录重命名 — 将
project/outputs/重命名为project/exports/,解决与会话输出目录的命名冲突。
下载
| 平台 | 下载链接 |
|---|---|
| macOS (Apple Silicon) | TorlyAI-2.6.1-arm64.dmg |
| macOS (Intel) | TorlyAI-2.6.1-x64.dmg |
| Windows (x64) | TorlyAI-Setup-2.6.1-x64.exe |
| Windows (ARM64) | TorlyAI-Setup-2.6.1-arm64.exe |
完整变更日志: v2.6.0...v2.6.1
TorlyAI Desktop v2.6.0
TorlyAI Desktop v2.6.0
Release Date: March 6, 2026
16 commits | 52 files changed | 3,585 insertions, 1,056 deletions
Highlights
- Portal Site & Patreon panels — New sidebar navigation for the project portal site and Patreon community, with redesigned WebsitePanel and PatreonCommunityPanel
- Living Business Plan performance — Added caching layer and debounced SSE/IPC events, significantly reducing dashboard re-render overhead
- Arch-specific Bun binaries — Each macOS DMG now bundles the correct Bun binary for its architecture (arm64/x64), fixing EBADARCH crashes on Intel Macs
- Agent file handling fixes — Deliverables now save to the correct session outputs path; OpenAI agent file writes aligned with session directory structure
- Build interceptor fix — All network interceptor dependencies are now correctly copied for ASAR unpacking, resolving missing-module errors in packaged builds
- Project exports refactor — Renamed
project/outputs/toproject/exports/to resolve naming collision with session outputs - Release tooling consolidation — Merged
/release-preflightinto pack commands; new/release-notescommand for post-upload finalization - VIP30 & Stripe live — Payment links cookbook added; Stripe credentials updated from test to live (innovatorly.ai account)
Full changelog: v2.5.0...v2.6.0
TorlyAI Desktop v2.6.0 — 发布说明
发布日期: 2026年3月6日
16 次提交 | 52 个文件变更 | 3,585 行新增,1,056 行删除
更新亮点
- 门户网站与 Patreon 面板 — 侧边栏新增项目门户网站和 Patreon 社区导航入口,重新设计了网站面板和社区面板
- 商业计划书实时更新性能优化 — 新增缓存层并对 SSE/IPC 事件进行防抖处理,大幅减少仪表盘重复渲染开销
- 架构匹配的 Bun 二进制文件 — macOS DMG 安装包现在为 arm64 和 x64 分别打包对应架构的 Bun 运行时,修复了 Intel Mac 上的 EBADARCH 崩溃问题
- 代理文件处理修复 — 交付物现在正确保存到会话输出路径;OpenAI 代理的文件写入路径与会话目录结构对齐
- 构建拦截器修复 — 所有网络拦截器依赖文件现在正确复制到 ASAR 解包目录,解决了打包后缺少模块的错误
- 项目导出重构 — 将
project/outputs/重命名为project/exports/,解决与会话输出目录的命名冲突 - 发布工具整合 — 将
/release-preflight合并到构建命令中;新增/release-notes命令用于上传后的发布收尾工作 - VIP30 与 Stripe 上线 — 新增支付链接操作手册;Stripe 凭证从测试环境更新为正式环境(innovatorly.ai 账户)
完整变更日志: v2.5.0...v2.6.0
TorlyAI Desktop v2.5.0
TorlyAI Desktop v2.5.0 — Free Tier, Status-Gated Consolidation & Multi-Backend Parity
Release Date: 2026-03-03
Highlights
- Free Tier Authorization — Try TorlyAI with 10 messages, no subscription required. Trial banner, upgrade modal, and feature gating built in
- Status-Gated BP Consolidation — Review and cherry-pick which agent outputs enter your Living Business Plan via a consolidation dialog
- Multi-Backend Agent Parity — ChatGPT/Codex sessions now have full skill injection, persona loading, and BP workflows identical to Claude sessions
New Features
- Free tier with
max_messages,max_sessions,max_connectionsenforcement TrialBannerwith 3 visual states (info → warning → danger)UpgradeModalwith plan-aware pricingConsolidateBPDialog— shared dialog for reviewing BP outputs before consolidation- Auto-trigger on session status change to Done/Closed
- Manual trigger via right-click → "Add Outputs to Business Plan"
- Consolidation queue with deduplication and sequential processing
- Auto-applies "BP Updated" label to consolidated sessions
- Consolidation mode toggle (auto vs gated) in Workspace Settings
Improvements
- Skill injection for non-Claude backends (SKILL.md loaded into system prompt)
- OpenAI
queryFnregistered at session creation time - Improved connection error messages (Pi SDK nested JSON extraction)
- Idempotent workspace activation with
current_workspace_iddefence-in-depth - Default permissions expanded (document CLI tools, typecheck scripts)
- Session list scroll fix (
h-screen→flex-1 min-h-0)
Bug Fixes
- Fixed interceptor module error on macOS Intel builds
- Fixed project tasks limited to 50 (removed hardcoded default)
- Disabled broken
torlyai-docsMCP server - Removed unshipped navigation entries
Documentation
- Status-Gated Consolidation PRD
- Data Storage Architecture Cookbook
- Workspaces-Sessions-Tasks Cookbook
- Free Tier Authorization PRD
- Authentication Cookbook Appendix A (archiving, multi-device)
- Architecture Doc Appendix F (Pi AI upgrade paths)
- Workspaces docs page rewritten on torly.ai/docs
Full changelog: 84 files changed, 8,071 insertions, 313 deletions across 13 commits.
Built with TorlyAI Desktop and Claude Opus 4.6.
TorlyAI Desktop v2.1.0
TorlyAI Desktop v2.1.0
Free Tier, Multi-LLM Support, and upstream v0.6.0 integration — the most feature-rich release yet.
Download
| Platform | File |
|---|---|
| macOS (Apple Silicon) | TorlyAI-arm64.zip |
| macOS (Intel) | TorlyAI-x64.zip |
| Windows x64 (Installer) | TorlyAI-Setup-2.1.0-x64.exe |
| Windows ARM64 (Installer) | TorlyAI-Setup-2.1.0-arm64.exe |
New Features
Free Tier & Trial System
Try TorlyAI without a subscription. The Free tier gives new users a taste of the full experience before committing.
- 10-message trial per session with real-time countdown in the chat input
- Trial banner with 3 visual states — info (plenty left), warning (running low), danger (nearly exhausted)
- Upgrade modal with plan-aware pricing — Free users see "Custom" for Autopilot; Premium users see the actual price
- Feature gating — Free tier has access to basic chat; Premium unlocks unlimited chat, Learning Center, PDF/DOCX export, Score Dashboard; Autopilot adds Kanban, Autopilot engine, and Project Website
- Usage tracking — client-side enforcement with server-synced JWT claims (
max_messages,max_sessions,max_connections) - Server-side JWT integration — Free tier limits are signed into the license token at issuance and refresh, with heartbeat returning usage stubs
Multi-LLM Provider Support
Connect your own AI providers beyond Claude.
- OpenAI-compatible backend — works with OpenAI, Azure OpenAI, and any OpenAI-compatible API
- Qwen CLI OAuth — authenticate with Alibaba Cloud's Qwen models
- Pi Agent enhancements — improved subprocess management and packaged build support
- Provider cap enforcement — Free tier limited to 1 provider; Premium/Autopilot unlimited
Upstream v0.6.0 Integration
Incorporates the latest from Craft Agents OSS:
- Built-in browser — in-app browser panes for web automation, form filling, and screenshots
- Document tools — PDF, Word, Excel, PowerPoint, and image processing built-in (markitdown, pdf-tool, xlsx-tool, docx-tool, pptx-tool, img-tool)
- Multi-panel layout — browser, documents, and conversations side by side
- Session branching — fork conversations to explore different approaches
- Hooks → Automations rebrand — event-driven hooks now called Automations with updated UI
- Sonnet 4.6 — automatic migration from Sonnet 4.5
Enhanced Onboarding
- Usage statistics in completion step — shows workspace metrics after setup
- Improved i18n — better Chinese localisation coverage
- Free tier onboarding flow — tailored experience for trial users
Project Workspace Improvements
- Project read tools — new IPC channels for reading project data
- Default skills bundle — pre-loaded skills for new workspaces
- AI task improvements — better natural language task parsing
Bug Fixes
- Banner layout fix — TrialBanner no longer overlaps panel headers; uses CSS variable
--torlyai-banner-hfor cross-component coordination - Windows invisible window — resolved 3 bugs preventing the app window from appearing on Windows (Bun path, try-catch isolation, afterPack name)
- macOS invisible window — fixed Bun subprocess path resolution in packaged builds
- OAuth callback — improved auth callback handling and license validation flow
- API endpoint trailing slashes — fixed TorlyAI API endpoints to use correct trailing slash format
- Asar unpacked path — runtime resolver now correctly finds files in
.asar.unpackedfor packaged builds - electron-builder OOM — resolved out-of-memory errors during macOS packaging
Build & Distribution
- Windows ARM64 — native ARM64 installer for Windows on ARM
- Improved build resilience — better error handling and validation in build pipeline
- Windows bash bundling — Bun subprocess works correctly on Windows
- Dual-repo upload — release artifacts published to both
torlyai/desktopandtorlyai/Torlyai-bp-master
Documentation
- 6 new PRDs including Free Tier Authorization, Pi Agent Server, and LLM Providers
- Updated architecture docs with upstream merge assessment
- macOS and Windows packaging postmortems
- Updated Terms of Service and Privacy Policy for BYOK pricing model
Server Changes (torly.ai)
- JWT signing now includes Free tier claims (
max_messages=10,max_sessions=1,max_connections=1) - Token refresh endpoint mirrors Free tier injection
- Heartbeat returns
usagestub for client-side enforcement - Deployed to production at torly.ai
Full Changelog: torlyai/Torlyai-bp-master@v2.0.0...v2.1.0
TorlyAI Desktop v1.7.0
TorlyAI Desktop v1.7.0
The biggest release yet — Setup Wizard, Autopilot, Authentication, and a redesigned Matrix Canvas.
Download
| Platform | File |
|---|---|
| macOS (Apple Silicon) | TorlyAI-1.7.0-arm64.dmg |
| macOS (Intel) | TorlyAI-1.7.0-x64.dmg |
| Windows (Installer) | TorlyAI-Setup-1.7.0.exe |
| Windows (Portable) | TorlyAI-1.7.0-win-x64.zip |
| Linux (AppImage) | TorlyAI-1.7.0.AppImage |
New Features
Setup Wizard
- 8-step guided onboarding — from workspace creation to first task, including profile setup, GDPR consent, 6-phase journey walkthrough, and AI team introduction
- Quick start actions — choose Idea Discovery, Seed Workflow (17-task template), or Jump to Chat
- Re-runnable from Settings → Re-run Setup Wizard
Autopilot Mode
- Autonomous agent orchestration — multi-session pipeline that runs agents in sequence
- Visual pipeline UI — monitor progress, pause/resume, and review outputs
- Smart routing — routes tasks to the right agent based on type and phase
Content Pipeline
- Living business plan assembly — agents publish section drafts that automatically flow into the master business plan
- Section mapping — each agent output maps to a specific BP section
- Version tracking — maintains revision history as content evolves
Authentication & Account System
- BYOK license system — bring your own API key with license validation
- Torly Account linking — connect your desktop app to your torly.ai account
- Subscription status — view plan details, usage, and billing in Account Settings
- Offline grace period — continue working offline with periodic license validation
- Heartbeat service — background license verification with graceful degradation
- Feature gating — tier-based access to premium features
Event-Driven Hooks
- 5 trigger events — LabelAdd, SessionStatusChange, PreToolUse, PostToolUse, SchedulerTick
- 2 action types — Command hooks (run shell commands) and Prompt hooks (inject AI instructions)
- Workspace-scoped — hooks configured per workspace for isolated automation
Task Dependencies & Acceptance Criteria
- Prerequisite chains — define blocking relationships between tasks
- Dependency-aware work queue — agents only pick up tasks whose prerequisites are complete
- Acceptance criteria — success rubrics per task with AI auto-generation via Claude
- Visual indicators — dependency badges on Kanban cards
Chinese Language Support
- Simplified Chinese (zh-CN) full UI localisation covering menus, settings, and all workspace pages
- Instant language toggle — switch between English and Chinese without restarting
Multi-Backend LLM Support
- Anthropic Claude, OpenAI/Codex, GitHub Copilot, Google AI Studio, OpenRouter, Ollama
- Searchable model picker with Best/Balanced/Fast tiers
- MCP sources proxied to non-Anthropic backends
Improvements
Redesigned Matrix Canvas
- Refactored canvas blocks, grid, and segments with improved layout and interactions
- New Criteria Sidebar for assessment criteria display
- New Exploration Hub for guided exploration entry points
- New Segment Progress Ring for visual progress indicators
- Refined animations and progress tracking
Enhanced DOCX Export
- Professional document formatting with improved styles and structure
- Table support with proper cell formatting
- Section structure matching business plan layout
- Improved typography and visual hierarchy
Score Dashboard
- Continuous assessment orchestrator with gap-to-task generation
- Automated quality feedback loop triggered by content changes
UI & Settings
- New Account Settings page with subscription management
- Improved Offline Grace Banner with countdown messaging
- Enhanced Onboarding Wizard with refined step flow and validation
- New settings icons and registry updates
- Updated sidebar navigation
Other Improvements
- File Previews — HTML inline rendering and data table export (XLSX/CSV)
- Keyboard shortcuts — Cmd+/ shortcut panel with focus zones
- Session search — ripgrep-based full-text search across all sessions
- Web-fetch SSRF hardening and Explore mode shell escape fixes
Auto-Update
Existing TorlyAI Desktop users will be prompted to update automatically. You can also check manually via Help → Check for Updates.
Full documentation: torly.ai/docs
Download page: torly.ai/download