Add Write File Tools to WaveAI (#2492)

also updates ROADMAP.md, and fixes a node pruning bug on the FE, and
adds a new diff viewer that we can view the write_text_file and
edit_text_file diffs in. adds a backup file system that can be used to restore AI edited files back to their original states.

Author: sawka

.roo/rules/rules.md

@@ -35,6 +35,7 @@ It has a TypeScript/React frontend and a Go backend. They talk together over `ws
     - Import the "cn" function from "@/util/util" to do classname / clsx class merge (it uses twMerge underneath)
     - For element variants use class-variance-authority
     - Do NOT create private fields in classes (they are impossible to inspect)
+    - Use PascalCase for global consts at the top of files
   - **Component Practices**:
     - Make sure to add cursor-pointer to buttons/links and clickable items
     - NEVER use cursor-help (it looks terrible)
@@ -48,6 +49,12 @@ It has a TypeScript/React frontend and a Go backend. They talk together over `ws
 - _never_ use cursor-help, or cursor-not-allowed (it looks terrible)
 - We have custom CSS setup as well, so it is a hybrid system. For new code we prefer tailwind, and are working to migrate code to all use tailwind.
 
+### RPC System
+
+To define a new RPC call, add the new definition to `pkg/wshrpc/wshrpctypes.go` including any input/output data that is required. After modifying wshrpctypes.go run `task generate` to generate the client APIs.
+
+For normal "server" RPCs (where a frontend client is calling the main server) you should implement the RPC call in `pkg/wshrpc/wshserver.go`.
+
 ### Code Generation
 
 - **TypeScript Types**: TypeScript types are automatically generated from Go types. After modifying Go types in `pkg/wshrpc/wshrpctypes.go`, run `task generate` to update the TypeScript type definitions in `frontend/types/gotypes.d.ts`.

ROADMAP.md

@@ -6,59 +6,82 @@ Want input on the roadmap? Join the discussion on [Discord](https://discord.gg/X
 
 Legend: ✅ Done | 🔧 In Progress | 🔷 Planned | 🤞 Stretch Goal
 
-## v0.11.0
-
-Released on 1/25/25
-
-- ✅ File/Directory Preview improvements
-  - ✅ Reworked fileshare layer running over RPC
-  - ✅ Expanded URI types supported by `wsh file ...`
-  - ✅ EC-TIME timeout when transferring large files
-- ✅ Fixes for reducing 2FA requests on connect
-- ✅ WebLinks in the terminal working again
-- ✅ Search in Web Views
-- ✅ Search in the Terminal
-- ✅ Custom init files for widgets and terminal blocks
-- ✅ Multi-Input between terminal blocks on the same tab
-- ✅ Gemini AI support
-- ✅ Various Connection Bugs + Improvements
-- ✅ More Connection Config Options
-
-## v0.11.1
-
-Targeting 1/31/25
-
-- 🔧 Reduce main-line 2FA requests to 1 per connection
-- 🔧 Remote S3 bucket browsing (directory + files)
-- 🔷 Drag & drop between preview blocks
-- 🔷 Drag into/out of a preview block from native file explorer
-- 🔷 Wave Apps (Go SDK)
-- 🔷 JSON schema support (basic)
-- 🤞 Frontend Only Widgets, React + Babel Transpiling in an iframe/webview
-
-## v0.12
-
-Targeting mid-February.
-
-- 🔷 Import/Export Tab Layouts and Widgets
-- 🔷 log viewer
-- 🔷 binary viewer
-- 🔷 New layout actions (splitting, replacing blocks)
-- 🔷 Rewrite of window/tab system
-- 🔷 Minimized / Non-Visible blocks
-- 🔷 Custom keybindings to quickly switch / invoke built-in and custom widgets
-- 🔷 More Drag & Drop support of files/URLs to create blocks
-- 🔷 Tab Templates
-
-## Planned (Unscheduled)
-
-- 🔷 Customizable Keybindings
-  - 🔷 Launch widgets with custom keybindings
-  - 🔷 Re-assign system keybindings
+## Current AI Capabilities
+
+Wave Terminal's AI assistant is already powerful and continues to evolve. Here's what works today:
+
+### AI Provider Support
+
+- ✅ OpenAI (including gpt-5 and gpt-5-mini models)
+
+### Context & Input
+
+- ✅ Widget context integration - AI sees your open terminals, web views, and other widgets
+- ✅ Image and document upload - Attach images and files to conversations
+- ✅ Local file reading - Read text files and directory listings on local machine
+- ✅ Web search - Native web search capability for current information
+- ✅ Shell integration awareness - AI understands terminal state (shell, version, OS, etc.)
+
+### Widget Interaction Tools
+
+- ✅ Widget screenshots - Capture visual state of any widget
+- ✅ Terminal scrollback access - Read terminal history and output
+- ✅ Web navigation - Control browser widgets
+
+## ROADMAP Enhanced AI Capabilities
+
+### AI Configuration & Flexibility
+
+- 🔷 BYOK (Bring Your Own Key) - Use your own API keys for any supported provider
+- 🔧 Enhanced provider configuration options
+- 🔷 Context (add markdown files to give persistent system context)
+
+### Expanded Provider Support
+
+Top priorities are Claude (for better coding support), and the OpenAI Completions API which will allow us to interface with
+many more local/open models.
+
+- 🔷 Anthropic Claude - Full integration with extended thinking and tool use
+- 🔷 OpenAI Completions API - Support for older model formats
+- 🤞 Google Gemini - Complete integration
+- 🤞 Local AI agents - Run AI models locally on your machine
+
+### Advanced AI Tools
+
+#### File Operations
+
+- 🔧 AI file writing with intelligent diff previews
+- 🔧 Rollback support for AI-made changes
+- 🔷 Multi-file editing workflows
+- 🔷 Safe file modification patterns
+
+#### Terminal Command Execution
+
+- 🔧 Execute commands directly from AI
+- 🔧 Intelligent terminal state detection
+- 🔧 Command result capture and parsing
+
+### Remote & Advanced Capabilities
+
+- 🔷 Remote file operations - Read and write files on SSH connections
+- 🔷 Custom AI-powered widgets (Tsunami framework)
+- 🔷 AI Can spawn Wave Blocks
+- 🔷 Drag&Drop from Preview Widgets to Wave AI
+
+### Wave AI Widget Builder
+
+- 🔷 Visual builder for creating custom AI-powered widgets
+- 🔷 Template library for common AI workflows
+- 🔷 Rapid prototyping and iteration tools
+
+## Other Platform & UX Improvements (Non AI)
+
+- 🔷 Import/Export tab layouts and widgets
+- 🔧 Enhanced layout actions (splitting, replacing blocks)
+- 🔷 Extended drag & drop for files/URLs
+- 🔷 Tab templates for quick workspace setup
+- 🔷 Advanced keybinding customization
+  - 🔷 Widget launch shortcuts
+  - 🔷 System keybinding reassignment
 - 🔷 Command Palette
-- 🔷 AI Context
-- 🔷 Monaco Theming
-- 🔷 File system watching for Preview
-- 🔷 File system watching for drag and drop
-- 🤞 Explore VSCode Extension Compatibility with standalone Monaco Editor (language servers)
-- 🤞 VSCode File Icons in Preview
+- 🔷 Monaco Editor theming

cmd/server/main-server.go

@@ -18,6 +18,7 @@ import (
 	"github.com/wavetermdev/waveterm/pkg/authkey"
 	"github.com/wavetermdev/waveterm/pkg/blockcontroller"
 	"github.com/wavetermdev/waveterm/pkg/blocklogger"
+	"github.com/wavetermdev/waveterm/pkg/filebackup"
 	"github.com/wavetermdev/waveterm/pkg/filestore"
 	"github.com/wavetermdev/waveterm/pkg/panichandler"
 	"github.com/wavetermdev/waveterm/pkg/remote/conncontroller"
@@ -55,6 +56,8 @@ const TelemetryTick = 2 * time.Minute
 const TelemetryInterval = 4 * time.Hour
 const TelemetryInitialCountsWait = 5 * time.Second
 const TelemetryCountsInterval = 1 * time.Hour
+const BackupCleanupTick = 2 * time.Minute
+const BackupCleanupInterval = 4 * time.Hour
 
 var shutdownOnce sync.Once
 
@@ -118,6 +121,23 @@ func telemetryLoop() {
 	}
 }
 
+func backupCleanupLoop() {
+	defer func() {
+		panichandler.PanicHandler("backupCleanupLoop", recover())
+	}()
+	var nextCleanup int64
+	for {
+		if time.Now().Unix() > nextCleanup {
+			nextCleanup = time.Now().Add(BackupCleanupInterval).Unix()
+			err := filebackup.CleanupOldBackups()
+			if err != nil {
+				log.Printf("error cleaning up old backups: %v\n", err)
+			}
+		}
+		time.Sleep(BackupCleanupTick)
+	}
+}
+
 func panicTelemetryHandler(panicName string) {
 	activity := wshrpc.ActivityUpdate{NumPanics: 1}
 	err := telemetry.UpdateActivity(context.Background(), activity)
@@ -446,6 +466,7 @@ func main() {
 	go stdinReadWatch()
 	go telemetryLoop()
 	go updateTelemetryCountsLoop()
+	go backupCleanupLoop()
 	go startupActivityUpdate(firstLaunch) // must be after startConfigWatcher()
 	blocklogger.InitBlockLogger()
 	go wavebase.GetSystemSummary() // get this cached (used in AI)

docs/docs/config.mdx

@@ -64,6 +64,7 @@ wsh editconfig
 | editor:stickyscrollenabled           | bool     | enables monaco editor's stickyScroll feature (pinning headers of current context, e.g. class names, method names, etc.), defaults to false                                                                                                                    |
 | editor:wordwrap                      | bool     | set to true to enable word wrapping in the editor (defaults to false)                                                                                                                                                                                         |
 | editor:fontsize                      | float64  | set the font size for the editor (defaults to 12px)                                                                                                                                                                                                           |
+| editor:inlinediff                    | bool     | set to true to show diffs inline instead of side-by-side, false for side-by-side (defaults to undefined which uses Monaco's responsive behavior)                                                                                                              |
 | preview:showhiddenfiles              | bool     | set to false to disable showing hidden files in the directory preview (defaults to true)                                                                                                                                                                      |
 | markdown:fontsize                    | float64  | font size for the normal text when rendering markdown in preview. headers are scaled up from this size, (default 14px)                                                                                                                                        |
 | markdown:fixedfontsize               | float64  | font size for the code blocks when rendering markdown in preview (default is 12px)                                                                                                                                                                            |

frontend/app/aipanel/aimessage.tsx

@@ -11,7 +11,15 @@ import { WaveUIMessage, WaveUIMessagePart } from "./aitypes";
 import { WaveAIModel } from "./waveai-model";
 
 const AIThinking = memo(
-    ({ message = "AI is thinking...", reasoningText }: { message?: string; reasoningText?: string }) => {
+    ({
+        message = "AI is thinking...",
+        reasoningText,
+        isWaitingApproval = false,
+    }: {
+        message?: string;
+        reasoningText?: string;
+        isWaitingApproval?: boolean;
+    }) => {
         const scrollRef = useRef<HTMLDivElement>(null);
 
         useEffect(() => {
@@ -30,17 +38,21 @@ const AIThinking = memo(
         return (
             <div className="flex flex-col gap-1">
                 <div className="flex items-center gap-2">
-                    <div className="animate-pulse flex items-center">
-                        <i className="fa fa-circle text-[10px]"></i>
-                        <i className="fa fa-circle text-[10px] mx-1"></i>
-                        <i className="fa fa-circle text-[10px]"></i>
-                    </div>
+                    {isWaitingApproval ? (
+                        <i className="fa fa-clock text-base text-yellow-500"></i>
+                    ) : (
+                        <div className="animate-pulse flex items-center">
+                            <i className="fa fa-circle text-[10px]"></i>
+                            <i className="fa fa-circle text-[10px] mx-1"></i>
+                            <i className="fa fa-circle text-[10px]"></i>
+                        </div>
+                    )}
                     {message && <span className="text-sm text-gray-400">{message}</span>}
                 </div>
                 {displayText && (
                     <div
                         ref={scrollRef}
-                        className="text-sm text-gray-500 overflow-y-auto max-h-[2lh] max-w-[600px] pl-9"
+                        className="text-sm text-gray-500 overflow-y-auto max-h-[3lh] max-w-[600px] pl-9"
                     >
                         {displayText}
                     </div>
@@ -172,7 +184,7 @@ const getThinkingMessage = (
     parts: WaveUIMessagePart[],
     isStreaming: boolean,
     role: string
-): { message: string; reasoningText?: string } | null => {
+): { message: string; reasoningText?: string; isWaitingApproval?: boolean } | null => {
     if (!isStreaming || role !== "assistant") {
         return null;
     }
@@ -182,7 +194,7 @@ const getThinkingMessage = (
     );
 
     if (hasPendingApprovals) {
-        return { message: "Waiting for Tool Approvals..." };
+        return { message: "Waiting for Tool Approvals...", isWaitingApproval: true };
     }
 
     const lastPart = parts[parts.length - 1];
@@ -214,7 +226,9 @@ export const AIMessage = memo(({ message, isStreaming }: AIMessageProps) => {
             <div
                 className={cn(
                     "px-2 rounded-lg [&>*:first-child]:!mt-0",
-                    message.role === "user" ? "py-2 bg-accent-800 text-white max-w-[calc(100%-20px)]" : null
+                    message.role === "user"
+                        ? "py-2 bg-accent-800 text-white max-w-[calc(100%-20px)]"
+                        : "min-w-[min(100%,500px)]"
                 )}
             >
                 {displayParts.length === 0 && !isStreaming && !thinkingData ? (
@@ -232,7 +246,11 @@ export const AIMessage = memo(({ message, isStreaming }: AIMessageProps) => {
                         )}
                         {thinkingData != null && (
                             <div className="mt-2">
-                                <AIThinking message={thinkingData.message} reasoningText={thinkingData.reasoningText} />
+                                <AIThinking
+                                    message={thinkingData.message}
+                                    reasoningText={thinkingData.reasoningText}
+                                    isWaitingApproval={thinkingData.isWaitingApproval}
+                                />
                             </div>
                         )}
                     </>