v1.5.2

Bowow — BETA Stable Version

Bug Fixes

• Fixed the settings window
• Fixed persistent auto-scroll to bottom
• Fixed an API error
• Fixed the HTML browser
• Fixed Markdown rendering
• Fixed the OpenRouter API
• Fixed MCP Tools
• Fixed the chat appearing centered on screen
• Fixed the Scroll indicators
• Fixed the electron-updater

---

New Additions

• Added an agent
• Added localEffort

All additions made so far (since the beginning)

Critical Fixes

• handleRevert data loss — read the original file from disk when no timeline entry exists
• Permission checks — assert() before create_file (overwrite) and edit_file
• System prompt — added "ask before modifying" to buildSystemPrompt
• Permission modal styling — removed the 🔐 icon, single line, centered #888 buttons, no overlay/border, shimmer effect similar to "Thinking"
• Permission scope — every BuildAgent now filters events by detail.sessionID !== String(buildId)
• Multi-build persistence — scheduleBuildSave saves all builds under build-agent-conv-all; loadAllSavedConversations() restores all of them
• Per-pane model picker — localModel state (instead of the global setApiModel)
• Diff preview — increased slice limits from 2000/4000 to 100,000 characters; ColoredDiff maxLines raised from 16 to 50
• Azure streaming — changed streamAzure() from res.json() (single blob) to an SSE line-by-line reader
• Electron-only fallbacks — LSP.start() and MCP.connectStdio() now use console.warn + return instead of throwing
• Google maxTokens — increased default from 4096 to 8192
• Loop detection threshold — repeatSigCount raised from 3 to 5
• selectElements stub — added selectedElements: string[] and made () => {} a real function
• useCommandPermission stub — requestPermission now actually waits for the user (real Promise)

Tool System Unification

• Registry tools renamed — read → read_file, edit → edit_file, write → create_file, bash → run_command, glob → glob_search, grep → grep_search
• Dynamic getToolDefs() — replaced the 480-line static TOOL_DEFS with generation from toolRegistry.toToolDefs() + a fallback map
• registerBuiltins() — moved to module level
• Removed broken AbsFile import — the file didn't actually exist
• Bridge at the top of execTool — passes buildId + apiSettings in context; removed duplicated inline handlers (task, glob_search, grep_search, create_file, read_file, run_command, edit_file)
• Removed dead code — validatePath, ALLOWED_NO_EXT, processInline, withFileMutationQueue, applyEditsToNormalizedContent, stripBom, detectLineEnding, normalizeToLF, restoreLineEndings

UI & Markdown

• Full Markdown support — replaced the custom processInline with react-markdown (headers, lists, code blocks, bold, italic)
• Collapse arrows — all elements collapsed by default; user messages expanded by default (the one exception)
• Arrow hidden for run_command only — read_file now has arrows (needed for HTML iframes)
• HTML iframe display — fullContent now includes read_file in addition to create_file and edit_file
• Auto-scroll fix — setUserScrolledUp(false) when a user message is added; setTimeout(0) in handleClear
• (+n)/(-n) stat indicators — enlarged from var(--font-xxs) to var(--font-sm)
• Unified background — all backgrounds now #121212
• Custom fonts — AR.otf and EN.otf added under public/font/, loaded via @font-face in index.css
• Inline code colors — background: '#121212', color: '#00adb5'
• Removed whiteSpace: 'pre-wrap' from the wrapper to fix vertically-broken text

Testing

• Vitest setup — vitest.config.ts + scripts added to package.json
• 16 tests across 3 files (permission, tools, registry)

Current Session

• Long text handling — wordBreak: 'break-word', overflowWrap: 'anywhere', inline code set to display: 'inline' + whiteSpace: 'nowrap'
• Silent editing — stopped explaining each step out loud
• Per-split-screen effort — localEffort state (independent per pane)
• Babel fix — added parentheses around (globalApiSettings.thinkingEffort || 'default')

v1.5.1

Bowow — BETA Stable Version

Bug Fixes

• Fixed the settings window
• Fixed persistent auto-scroll to bottom
• Fixed an API error
• Fixed the HTML browser
• Fixed Markdown rendering
• Fixed the OpenRouter API
• Fixed MCP Tools
• Fixed the chat appearing centered on screen

---

New Additions

• Added an agent
• Added localEffort

All additions made so far (since the beginning)

Critical Fixes

• handleRevert data loss — read the original file from disk when no timeline entry exists
• Permission checks — assert() before create_file (overwrite) and edit_file
• System prompt — added "ask before modifying" to buildSystemPrompt
• Permission modal styling — removed the 🔐 icon, single line, centered #888 buttons, no overlay/border, shimmer effect similar to "Thinking"
• Permission scope — every BuildAgent now filters events by detail.sessionID !== String(buildId)
• Multi-build persistence — scheduleBuildSave saves all builds under build-agent-conv-all; loadAllSavedConversations() restores all of them
• Per-pane model picker — localModel state (instead of the global setApiModel)
• Diff preview — increased slice limits from 2000/4000 to 100,000 characters; ColoredDiff maxLines raised from 16 to 50
• Azure streaming — changed streamAzure() from res.json() (single blob) to an SSE line-by-line reader
• Electron-only fallbacks — LSP.start() and MCP.connectStdio() now use console.warn + return instead of throwing
• Google maxTokens — increased default from 4096 to 8192
• Loop detection threshold — repeatSigCount raised from 3 to 5
• selectElements stub — added selectedElements: string[] and made () => {} a real function
• useCommandPermission stub — requestPermission now actually waits for the user (real Promise)

Tool System Unification

• Registry tools renamed — read → read_file, edit → edit_file, write → create_file, bash → run_command, glob → glob_search, grep → grep_search
• Dynamic getToolDefs() — replaced the 480-line static TOOL_DEFS with generation from toolRegistry.toToolDefs() + a fallback map
• registerBuiltins() — moved to module level
• Removed broken AbsFile import — the file didn't actually exist
• Bridge at the top of execTool — passes buildId + apiSettings in context; removed duplicated inline handlers (task, glob_search, grep_search, create_file, read_file, run_command, edit_file)
• Removed dead code — validatePath, ALLOWED_NO_EXT, processInline, withFileMutationQueue, applyEditsToNormalizedContent, stripBom, detectLineEnding, normalizeToLF, restoreLineEndings

UI & Markdown

• Full Markdown support — replaced the custom processInline with react-markdown (headers, lists, code blocks, bold, italic)
• Collapse arrows — all elements collapsed by default; user messages expanded by default (the one exception)
• Arrow hidden for run_command only — read_file now has arrows (needed for HTML iframes)
• HTML iframe display — fullContent now includes read_file in addition to create_file and edit_file
• Auto-scroll fix — setUserScrolledUp(false) when a user message is added; setTimeout(0) in handleClear
• (+n)/(-n) stat indicators — enlarged from var(--font-xxs) to var(--font-sm)
• Unified background — all backgrounds now #121212
• Custom fonts — AR.otf and EN.otf added under public/font/, loaded via @font-face in index.css
• Inline code colors — background: '#121212', color: '#00adb5'
• Removed whiteSpace: 'pre-wrap' from the wrapper to fix vertically-broken text

Testing

• Vitest setup — vitest.config.ts + scripts added to package.json
• 16 tests across 3 files (permission, tools, registry)

Current Session

• Long text handling — wordBreak: 'break-word', overflowWrap: 'anywhere', inline code set to display: 'inline' + whiteSpace: 'nowrap'
• Silent editing — stopped explaining each step out loud
• Per-split-screen effort — localEffort state (independent per pane)
• Babel fix — added parentheses around (globalApiSettings.thinkingEffort || 'default')

---

Important Note (Beta)

The program has full permissions and executes commands without asking for approval.

v1.5.0

Bowow - BETA stable version

Bugfixes

• Fixed writeBuildFile
• Fixed persist middleware.
• Fixed localStorage & localStorage.setItem
• Fixed useAppStore(s => s.builds)
• Fixed MarkdownContent
• Fixed contentLoaded

Important Note (Beta):

The program has full permissions and executes commands without asking for approval.

v1.0.0

Bowow - BETA

Important: This build is currently unstable. There is a known issue with the persist middleware. A fix is underway; please avoid downloading this version for now."

open source AI coding agent

Usage

• Launch the app and open the Settings panel to configure your AI provider and API key, and click done.
• Connect to a model (Gemini, OpenAI, Ollama, etc.).
• Open a project folder ( /new ) or start a new session.
• Describe what you want to build — the agent will generate files, edit code, and run commands.
• Use F10 to toggle split-screen mode for multi-session management.
• For close split-screen Use F10 not (X)

Full support for sending and receiving all file types and images

Important Note (Beta):

The program has full permissions and executes commands without asking for approval.

Be careful with your input: Files may be deleted if you ask for it, or if the context suggests