Back to docs

Changelog

2026-10-07

Changelog

v3.6.5 — 2026-10-05

Release date: 2026-10-05. Source: the releaseDate: '2026-10-05T13:38:15.614Z' of https://files.doforu.ai/downloads/latest-mac.yml and the releaseDate: '2026-10-05T14:33:14.880Z' of latest.yml (both verified by direct request). Version number basis: DESKTOP_APP_VERSION = "3.6.5" in src/config/version.ts (bumped from 3.6.0 to 3.6.5 in commit 4e216ba on 2026-10-05). This entry only includes changes traceable to repository files/commits or online artifacts. Desktop 3.6.2 / 3.6.3 are internal version numbers, 3.6.4 never existed (see "Version Gap Notes" below), and 3.6.1 is covered in the next section.

Release & Distribution

  • Site version number aligned to 3.6.5: DESKTOP_APP_VERSION was bumped from 3.6.0 to 3.6.5, and the macOS arm64 links on the download page and the homepage download section were changed to the canonical name with -arm64 (Source: commit 4e216ba).
  • Standardized release artifact naming: clarified the artifact naming rules for each platform (macOS Doforu-<v>-<arch>.dmg/.zip, Windows Doforu Setup <v>.exe, Linux Doforu-<v>.AppImage / doforu_<v>_amd64.deb / doforu-<v>.x86_64.rpm), and updated install.sh / install.ps1 and the site consumers accordingly (Source: docs/RELEASE-ARTIFACT-NAMING.md, commit 2e99657; for the naming fix commit contents see 4e216ba).
  • Added Linux to the download page: added an AppImage primary button and .deb / .rpm secondary links, and changed the card grid to md:grid-cols-2 lg:grid-cols-4 (Source: commit 1c1555c).
  • Install script usability fixes: fixed defects in install.sh / install.ps1 on Linux / macOS / Windows, keeping all four scripts (including the public/ copies) in sync (Source: commits c485640, e6249a8).
  • Version resolution chain rework: the install scripts now use a three-tier resolution of "CDN latest-*.yml → verify the artifact by platform → local fallback constant DEFAULT_VERSION", removing the defunct GitHub API fallback and converging the fallback version number into a single constant (Source: docs/RELEASE-ARTIFACT-NAMING.md section 5, commit 37f8b9c).

Models & Copy

  • Supported model list update: added the Claude Opus 5.5, Gemini 4 Argon, DeepSeek V4.1, Qwen3.8-Max, MiniMax M3 and other model families (Source: changelogItems in src/config/locale/messages/en/downloadPage.json, commit 14570a4).
  • Unified the speedup wording to 16x (previously 10x), see downloadPage.json and stats.json (Source: commit 14570a4).
  • Unified the site-wide slogan to "Give it a goal" ("给它一个目标"), and corrected the app category classification (Source: commit 7493c95).

Site & Security

  • Added a license ownership probe GET /api/billing/license/mine (returns only a boolean, without leaking the key), so purchased users no longer see the "Upgrade to Pro" entry point; added src/shared/lib/license-token.ts to add an RS256 signature token to the local license to prevent local database tampering (Source: commits 00c48dc, 17a6db0).
  • Completed the forgot password / reset password flow, with the email Logo asset domain fixed to https://doforu.ai (Source: commit 43760e4).
  • Dead code cleanup: removed unreferenced AI generator, chat components, session context and testimonials components along with their i18n copy (Source: commits b389db6, 328e24b, 07195b0, 14570a4).

Published Artifacts (Verified)

The 3.6.5 artifacts on https://files.doforu.ai/downloads (verified by direct request, HTTP 200):

Platform Artifact
macOS arm64 Doforu-3.6.5-arm64.dmg, Doforu-3.6.5-arm64.zip
macOS x64 Doforu-3.6.5-x64.dmg
Windows x64 Doforu Setup 3.6.5.exe
Linux Doforu-3.6.5.AppImage, doforu_3.6.5_amd64.deb, doforu-3.6.5.x86_64.rpm

Version Gap Notes

  • 3.6.4 never existed. The desktop repository's full history (all refs, git log --grep="3.6.4", git log -S"3.6.4") has zero hits, and there are no artifacts online either. The previous wording of "no changelog found" was not accurate enough; it should be judged directly as non-existent.
  • 3.6.2 / 3.6.3 are internal version numbers, with no release evidence. The repository contains two version bump commits, chore: bump version to 3.6.2 and minor fixes (2026-06-09) and chore: bump version to 3.6.3, remove telemetry (2026-07-08), but there are no platform artifacts online, no git tag, and no record in weekly reports or release docs; moreover the subsequent mainline version number fell back to 3.5.6 (2026-08-02, along with the Vite migration). Accordingly they are not listed as released versions.
  • Verification boundary: the desktop main repository doforu-labs/Doforu is a private repository, so GitHub Releases cannot be verified via the anonymous API (returns 404). Therefore the judgment of "not released" rests on the three pieces of evidence of online artifacts + git tag + in-repository docs, rather than a complete enumeration of distribution channels.

v3.6.1 — 2026-08-05

Release date: 2026-08-05. Source: the desktop repository's docs/weekly-report-2026-08-08.md (the 2026-08-02 ~ 08-08 weekly report) explicitly records "v3.6.1 released + performance optimization (8/5)" and lists the changes item by item. Distribution channel note: no 3.6.1 artifacts were found on the online CDN; at the time the primary version resolution path of the install script was GitHub Releases (see api.github.com/.../releases/latest in the history of install.sh), and that repository is private, so the distribution record cannot be reviewed via a public interface.

Performance

  • First-paint performance optimization: modals now render on demand, html2canvas loads via dynamic import, and first-paint JS is reduced by 46% (2339 KB → 1258 KB).
  • Database layer replacement: replaced better-sqlite3 with Node's built-in node:sqlite, retired the native rebuild script, and removed the cross-platform recompile dependency.

Fixes

  • Fixed unresponsive theme / language switching.
  • Markdown rendering: fixed KaTeX formulas being stripped by rehype-sanitize (added a MathML whitelist); added a /markdown-test stateless rendering test page.

New Capabilities

  • Vision model parsing: added a "Use vision model" toggle; when read_file reads an image it is forcibly handed to a separate vision model for parsing, and vision calls are switched to OCR mode (prioritizing text extraction from the image).
  • Allowance and settings: when the free allowance is exceeded, an upgrade dialog pops up automatically; removed the advanced runtime settings panel (the related copy was merged into the subagent area).

Follow-up Iterations in the Same Cycle (8/6–8/7)

The following changes took place after the 8/5 release, within the cycle where the version number remained at 3.6.1 (it was not bumped to 3.6.5 until 2026-09-30):

  • Subagent card rebuild: cards changed to a square grid layout (4 per row), a green status indicator light, and full theme color following; the internal model thinking stream changed to real-time SSE push (dual-threshold throttling), with tool commands and the thinking stream displayed together; a single card completes independently (no longer waiting for the batch), and failed / canceled states show an error message instead of a leftover placeholder.
  • Lock-free superseding concurrency (architectural): removed the wait lock before creating a new run, switching to non-blocking superseding concurrency; the effective path for the superseded flag was completed, with superseded runs keeping the canceled terminal-state semantics; run-scoped msgOps isolation fixed cross-contamination when saving session messages.
  • SSE terminal states as events: frontend stats and allowance polling (50 ms / 3 s) changed to event-driven, and cancellation detection changed from polling to abort-event-driven.
  • Added an "Allow LAN access" setting (off by default, takes effect via hot update).
  • Cleanup: removed 14 zero-call HTTP routes, 6 frontend dead wrapper functions and dead constants.

v3.6.0 — 2026-06-03

Highlights

  • v3.6.0 official release: the latest stable version of the desktop client, with comprehensive enhancements to MCP stability and the Skills system.
  • MCP integration improvements: improved MCP server connection stability and added support for live configuration refresh without restarting the app.
  • Skills system enhancements: added Claude Code Skill compatibility and expanded AI coding capabilities further.

Installation & Distribution

  • Added a curl | sh one-line install script supporting macOS / Linux / Windows (Git Bash/WSL).
  • Added a PowerShell install script (install.ps1) for a more complete Windows installation experience.
  • Added version labels to the download page, with support for both macOS Apple Silicon and Intel architectures.

Docs & Community

  • Added VS Cursor and VS Claude Code comparison pages.
  • Added SEO improvements: hreflang tags, structured data, and a more complete sitemap.
  • Added Japanese (日本語) and Korean (한국어) README translations.

v3.5.0 — 2026-05-28

New Features

  • One-line install script: supports quick installation via curl -fsSL https://doforu.ai/install.sh | sh.
  • PowerShell installation support: added the Windows install.ps1 script.
  • macOS Intel build: added macOS x64 architecture support to the desktop CI build pipeline, covering Intel Mac users.

Improvements & Fixes

  • Optimized the electron-builder packaging configuration to reduce installer size.
  • Fixed multi-window crashes for improved cross-platform stability.
  • Upgraded the download page UI to distinguish between Apple Silicon and Intel versions.

v3.4.0 — 2026-05-18

SEO & Marketing

  • Comprehensive SEO improvements: fixed 73 page indexing issues in Google Search Console.
  • Added VS comparison pages (vs Cursor, vs Claude Code) with FAQ Schema structured data.
  • Multi-language SEO fixes: hreflang tags, dynamic html lang, and sitemap linkage.
  • Unified social share images (OG Image / Twitter Card).

Content & Docs

  • Fully updated the API Key configuration docs, with newly added models such as GPT-5.5, DeepSeek V4, and Qwen 3.
  • Added a Provider table component to improve how model configurations are displayed.
  • Optimized the layout and typography of documentation tables.

v3.3.0 — 2026-05-10

Admin Dashboard

  • Added an admin License management system: supports issuing, enabling, disabling, and deleting licenses.
  • Added support for anonymously gifting licenses (giveaway).
  • The License list displays associated information such as user email and name.
  • Added RBAC permission control: permission checks for License management APIs.

System Improvements

  • Added Docker build support (pnpm + multi-stage build).
  • Fixed session crosstalk across multiple windows.
  • Added database connection timeout retries and keepalive configuration.

v3.2.0 — 2026-04-28

Model Ecosystem Expansion

  • Added GPT-5.5: support for OpenAI's latest flagship model.
  • Added DeepSeek V4 / V4 Flash: strengthened code generation capabilities.
  • Added Qwen 3.6: the latest version of Alibaba's Tongyi Qianwen.
  • Added Claude Opus 4.7 / Sonnet 4.5: support for Anthropic's latest models.
  • OpenRouter integration: support for accessing a multi-model gateway through OpenRouter.
  • Model leaderboard update: reorganized the recommended priority of models from each provider.

Configuration Experience

  • Added hot switching between Providers, so AI models can be switched without restarting the app.
  • Optimized the API Key configuration page, with newly added support for custom Base URLs.
  • Added model category tags and search.

v3.1.0 — 2026-04-15

New Features

  • Skills enhancements: added Claude Code Skill compatibility support to further expand AI coding capabilities.
  • Configuration hot refresh: supports live refresh of Skills and MCP configurations in settings without restarting the app.
  • macOS Intel build: added macOS x64 architecture support to the desktop CI build pipeline.

Fixes & Improvements

  • Fixed an issue where opening a new window would unexpectedly load an old conversation.
  • Fixed the Shell tool password dialog icon and command display issues.
  • Fixed the before-pack script formatting and fs/promises import to improve packaging stability.
  • Optimized electron-builder compatibility and packaging size.

v3.0.0 — 2026-04-03

Core Architecture Upgrade

  • Context Compression Mechanism: Introduced context-compactor, replacing the previous pointer-based compression, unifying the background compression strategy, and optimizing token usage and response speed in long-conversation scenarios.
  • Memory Module Switch: Added the memoryEnabled configuration, supporting global enable/disable of the memory feature and controlling Embedding model packaging and startup preloading behavior.
  • BYOK Architecture: API Keys are now stored in the frontend localStorage, supporting Bring Your Own Key; also fixed the issue where an empty Key would overwrite the global configuration.

Stability & Performance

  • Multi-Window Isolation: Removed the global window-storage cache, fully resolving multi-window session cross-talk and Electron multi-window crashes.
  • Freeze Detection: Optimized connection keep-alive and freeze detection under background/sleep states.
  • Token Management Optimization: Unified the token statistics system, replacing the tokenizer with tiktoken, and prioritizing the usage field returned by the API.
  • Client Retry: Added client retry, timeout, and keepalive configurations, improving reliability in weak-network environments.

Sessions & UI

  • The sidebar session list now supports statistics display and delete operations.
  • Hid the Session ID display on the homepage, keeping the DevTools debug output.
  • Unified the use of api.json for configuration management, removing redundant custom variables from .env.

v2.7.0 — 2026-03-18

Agents & Subagents

  • Multiple Todo tasks can now be marked as in_progress at the same time, improving parallel task visualization.
  • Added subagent duration display to make it easier to track the execution efficiency of parallel agents.
  • Prohibited nested subagent calls and optimized the runId generation strategy to avoid recursion risks.
  • Fixed the subagent cancellation mechanism to distinguish between timeouts and user-initiated cancellation, eliminating false "Operation cancelled by user" reports.

Tools & Interaction

  • Replaced search-replace-tool with edit-file-tool, supporting semantic code editing.
  • Optimized ASCII art (wireframe) rendering, adding CJK-friendly font support.
  • Fixed abnormal switching between Agent/Orchestrator/Plan modes; MCP tools are disabled by default in Plan mode to improve security.
  • Optimized running-status hints and the typing-dots animation experience.

Architecture Changes

  • Completely removed legacy CLI code, switching uniformly to HTTP mode.
  • Improved the graceful shutdown logic: automatically cleaning up MCP connections, connection pools, and Shell subprocesses, with added protection against repeated triggering.
  • Default port changed from 30008 to 17890.

v2.6.5 — 2026-03-10

Initial Desktop Agent Release

  • Initial release of the Doforu local-first AI coding assistant desktop client.
  • Built-in Skills system and MCP integration support.
  • Supports both macOS and Windows platforms.

Core Features

  • Supports three execution modes: Plan, Agent (single agent), and Orchestrator.
  • Multi-Agent parallel execution architecture that automatically decomposes and distributes tasks.
  • The main Agent automatically assesses complexity: lightweight tasks → fast subagent, deep tasks → extreme subagent.
  • A complete toolchain for filesystem read/write, command-line execution, web search, and more.

v2.6.0 — 2026-03-05

Feature Refinements

  • Optimized Electron multi-window support, fixing the window focus switching issue.
  • Added conversation history management and search.
  • Improved the stability of MCP tool calls, adding timeout detection.
  • Optimized memory usage, reducing resource consumption in idle states.

User Experience

  • Added dark mode support, following the system theme automatically.
  • Optimized the first-launch onboarding flow.
  • Improved the settings page layout and interaction.

v2.5.0 — 2026-02-25

Agent Capability Upgrade

  • Skills System Official Release: Supports creating, editing, and reusing custom AI capability modules.
  • Added the read_skill tool, supporting on-demand skill loading.
  • Added the create_skill / edit_skill APIs, supporting dynamic creation and modification of skills at runtime.
  • Skills support multilingual (Chinese/English) instruction templates.

Model Expansion

  • Added support for domestic models: ByteDance Doubao, Moonshot Kimi, Baidu ERNIE 5.0, iFlytek Spark, Zhipu GLM, Baichuan, and SenseTime SenseNova.
  • Added support for local models (Ollama / vLLM and other OpenAI-compatible interfaces).
  • Redesigned the model selector UI.

v2.4.0 — 2026-02-18

MCP Integration

  • Official MCP protocol support: implemented the Model Context Protocol standard interface.
  • Integrated external tools: database queries, web search, browser automation.
  • Automatic discovery and configuration of MCP servers.
  • Compatible with importing Cursor / Claude Desktop MCP configurations.

Tool Ecosystem

  • Support for both stdio and HTTP MCP transport modes.
  • Hot reload of MCP configuration, no restart required.
  • Added an MCP connection status monitoring and error diagnosis panel.

v2.3.0 — 2026-02-10

Plan Mode

  • Plan Mode release: outputs a complete execution blueprint after a requirement is entered (steps, tool selection, estimated time).
  • One-click switch to Agent or Orchestrator mode for execution after user confirmation.
  • Plan output supports visualized flowcharts.

UI Improvements

  • Redesigned the chat interface, supporting real-time display of the message stream.
  • Added a tool call card view: click to expand/collapse and inspect parameters and return values.
  • The bottom status bar shows the current mode and running status.
  • Added a "Stop" button to interrupt a task at any time.

v2.2.0 — 2026-02-01

Desktop Client Beta

  • Electron desktop client Beta release.
  • Native file system read/write capability (beyond browser sandbox limits).
  • Local command line execution support.
  • System tray and background running support.

Model Integrations

  • Support for OpenAI (GPT-5, o3 series).
  • Support for Anthropic Claude (Opus, Sonnet series).
  • Support for Google Gemini series.
  • BYOK (Bring Your Own Key) architecture design completed.

v2.1.0 — 2026-01-20

Multi-Agent Architecture

  • Orchestrator orchestration mode Alpha release: the main Agent automatically decomposes complex tasks, with multiple subagents executing in parallel.
  • Support for two subagent types: fast (fast and lightweight) and extreme (deep and complex).
  • Automatic collaboration and result aggregation among subagents.
  • Support for tool call chain tracing and debugging.

Framework Upgrade

  • Refactored the Electron main process architecture, supporting multi-window isolation.
  • The front-end UI framework transitioned from pure Web to a desktop-native experience.
  • Design and implementation of the inter-process communication (IPC) layer.

v2.0.0 — 2026-01-06

Major Pivot: From SaaS Template to Desktop Agent Platform

  • An entirely new product positioning: Doforu pivoted from an AI SaaS development framework to a Desktop AI Agent Orchestrator.
  • Discarded the template architecture and redesigned it as a desktop-native application.
  • Built the desktop app on Electron + Next.js.

First Desktop Release

  • Support for local AI assistant conversation.
  • Basic file read/write and code editing capability.
  • Multi-model API configuration and switching.
  • Local-first: all data is stored locally on the user's machine.

v1.0.1 — 2025-11-11

See the Doforu Two release log

  • Based on Next.js 16, with a significant improvement in compilation performance.
  • Replaced Auth.js with Better-Auth, supporting more authentication methods.
  • Admin dashboard UI improvements.
  • Support for visual configuration and RBAC permission control.
  • Integrated more complete content management features.

v1.0.0 — 2024-11-24

See the Doforu One release log

  • Initial release of Doforu.
  • A polished Landing Page + 20+ page components.
  • User authentication (Google / GitHub sign-in).
  • Data analytics (Google Analytics / OpenPanel).
  • i18n multi-language support.
  • Data storage (based on Supabase).
  • Payment system (based on Stripe).
  • Deployment support (Vercel / Cloudflare).