diff --git a/src/routes/api/write/+server.ts b/src/routes/api/write/+server.ts
index 1fd2f9c92..cc66f4f59 100644
--- a/src/routes/api/write/+server.ts
+++ b/src/routes/api/write/+server.ts
@@ -4,6 +4,10 @@ import Anthropic from '@anthropic-ai/sdk'
 
 // Safely access the API key, will be undefined if not set
 const ANTHROPIC_API_KEY_FOR_WRITE = env.ANTHROPIC_API_KEY_FOR_WRITE || undefined
+// NEW: Add global toggle for web search functionality
+const ENABLE_WEB_SEARCH = env.ENABLE_WEB_SEARCH !== 'false' // Default to true unless explicitly disabled
+// NEW: Add configurable rate limiting for tool calls per step
+const MAX_TOOL_CALLS_PER_STEP = parseInt(env.MAX_TOOL_CALLS_PER_STEP || '3')
 
 // Flag to track if API is available
 const IS_API_AVAILABLE = !!ANTHROPIC_API_KEY_FOR_WRITE
@@ -16,12 +20,86 @@ if (!IS_API_AVAILABLE) {
 }
 
 // Define step types for server-side use
-type StepName = 'research' | 'firstDraft' | 'firstCut' | 'firstEdit' | 'toneEdit' | 'finalEdit'
+type StepName =
+	| 'findTarget'
+	| 'webSearch'
+	| 'research'
+	| 'firstDraft'
+	| 'firstCut'
+	| 'firstEdit'
+	| 'toneEdit'
+	| 'finalEdit'
+
+// Define workflow types
+type WorkflowType = '1' | '2' | '3' | '4'
+
+// NEW: Define step configuration interface for tool usage
+interface StepConfig {
+	toolsEnabled?: boolean // Whether this step can use tools
+	maxToolCalls?: number // Maximum tool calls for this step (overrides global)
+	description?: string // Enhanced description when tools are used
+}
+
+// ENHANCED: Extend workflow configuration to support step configs
+type WorkflowConfig = {
+	steps: StepName[]
+	description: string
+	stepConfigs?: Record<StepName, StepConfig> // NEW: Optional step-level configuration
+}
+
+// NEW: Define step-level tool configurations
+const stepConfigs: Record<StepName, StepConfig> = {
+	// Research-focused steps that benefit from web search
+	findTarget: {
+		toolsEnabled: true,
+		maxToolCalls: 5,
+		description: 'Find possible targets (using web search)'
+	},
+	webSearch: {
+		toolsEnabled: true,
+		maxToolCalls: 3,
+		description: 'Research the target (using web search)'
+	},
+	research: {
+		toolsEnabled: false,
+		maxToolCalls: 2,
+		description: 'Auto-fill missing user inputs'
+	},
+	// Text processing steps remain tool-free for performance
+	firstDraft: { toolsEnabled: false },
+	firstCut: { toolsEnabled: false },
+	firstEdit: { toolsEnabled: false },
+	toneEdit: { toolsEnabled: false },
+	finalEdit: { toolsEnabled: false }
+}
 
+const workflowConfigs: Record<WorkflowType, WorkflowConfig> = {
+	'1': {
+		steps: ['findTarget'],
+		description: 'Find Target Only',
+		stepConfigs // NEW: Include step configurations
+	},
+	'2': {
+		steps: ['webSearch', 'research'],
+		description: 'Web Search + Autofill',
+		stepConfigs // NEW: Include step configurations
+	},
+	'3': {
+		steps: ['research'],
+		description: 'Autofill only',
+		stepConfigs // NEW: Include step configurations
+	},
+	'4': {
+		steps: ['firstDraft', 'firstCut', 'firstEdit', 'toneEdit', 'finalEdit'],
+		description: 'Full Email Generation',
+		stepConfigs // NEW: Include step configurations
+	}
+}
 // Server-side state management interface (not exposed to client)
 interface WriteState {
 	step: StepName | 'complete' | 'start' // Current/completed step
-	userInput: string // Original input from form
+	workflowType: WorkflowType // Type of workflow being executed
+	userInput: string // Original input from form (cleaned, without prefix)
 	email: string // Current email content
 	information?: string // Processed information after research
 	completedSteps: Array<{
@@ -52,6 +130,7 @@ const System_Prompts: { [id: string]: string } = {}
 System_Prompts['Basic'] = `You are a helpful AI assistant.
 
 Note: Some fields in the information may begin with a robot emoji (🤖). This indicates the field was automatically generated during research. You can use this information normally, just ignore the emoji marker.`
+
 System_Prompts['Mail'] = `
 What follows is the anatomy of a good email, a set of guidelines and
 criteria for writing a good mail. Each paragraph, except the last represents 
@@ -177,9 +256,41 @@ Example:
 Original: "Preferred communication style: undefined"
 Your output: "Preferred communication style: 🤖 Formal but approachable"
 
+Please remember that you are addressing this person, and try to make all inferences based on the information provided and your own knowledge. Err on the side of caution: if you are unsure, be polite and neutral.
+
 Output the full information, including your edits. Output nothing else.
 `
 
+System_Prompts['Target'] = `
+Please use your internet search capability to find individuals involved with AI safety who match the following description.
+
+For each person you find (aim for 3-5 people), please provide:
+1. Name and current position
+2. Why they're relevant to AI safety
+3. Their organization
+4. Brief note on their public stance on AI safety
+
+Please cite your sources for each person.
+`
+
+//Preface with '[Person's Name] = John Doe' etc.
+System_Prompts['webSearch'] = `
+Please use your internet search capability to research [Person's Name] who is [current role] at [organization/affiliation]. I plan to contact them about AI safety concerns.
+
+Search for and provide:
+1. Professional background (education, career history, notable positions)
+2. Their involvement with AI issues (policy positions, public statements, initiatives, articles, interviews)
+3. Their public views on AI development and safety (with direct quotes where possible)
+4. Recent activities related to technology policy or AI (last 6-12 months)
+5. Communication style and key terms they use when discussing technology issues
+6. Notable connections (organizations, committees, coalitions, or influential individuals they work with)
+7. Contact information (professional email or official channels if publicly available)
+
+Please cite all sources you use and only include information you can verify through your internet search. If you encounter conflicting information, note this and provide the most reliable source.
+
+BE BRIEF! This is extremely important. Try to output only a few lines of text for each questions.
+`
+
 // Only initialize the client if we have an API key
 const anthropic = IS_API_AVAILABLE
 	? new Anthropic({
@@ -191,13 +302,46 @@ export async function GET() {
 	return json({ apiAvailable: IS_API_AVAILABLE })
 }
 
-// Helper function to call Claude API with timing
+// Helper function to parse workflow type from user input
+function parseWorkflowType(userInput: string): {
+	workflowType: WorkflowType
+	cleanedInput: string
+} {
+	const workflowMatch = userInput.match(/^\[([1-4])\](.*)$/s)
+
+	if (workflowMatch) {
+		const workflowType = workflowMatch[1] as WorkflowType
+		const cleanedInput = workflowMatch[2].trim()
+		return { workflowType, cleanedInput }
+	}
+
+	// Default to workflow 4 if no prefix is found
+	return { workflowType: '4', cleanedInput: userInput }
+}
+// NEW: Interface for tool use response content
+interface ToolUseContent {
+	type: 'tool_use'
+	id: string
+	name: string
+	input: any
+}
+
+// NEW: Interface for tool result content
+interface ToolResultContent {
+	type: 'tool_result'
+	tool_use_id: string
+	content: string
+}
+
+// ENHANCED: Modified function signature to support optional tool usage
 async function callClaude(
 	stepName: string,
 	promptNames: string[],
-	userContent: string
+	userContent: string,
+	toolsEnabled: boolean = false // NEW: Optional parameter for tool usage
 ): Promise<{ text: string; durationSec: number }> {
 	const pencil = '✏️'
+	const search = '🔍' // NEW: Icon for tool usage
 	const logPrefix = `${pencil} write:${stepName}`
 	const startTime = Date.now()
 
@@ -212,73 +356,209 @@ async function callClaude(
 		// Combine all the specified prompts
 		const systemPrompt = promptNames.map((name) => System_Prompts[name]).join('')
 
-		const response = await anthropic.messages.create({
+		// NEW: Determine if tools should be included in this call
+		const shouldUseTools = toolsEnabled && ENABLE_WEB_SEARCH && IS_API_AVAILABLE
+
+		// NEW: Log tool usage status
+		if (shouldUseTools) {
+			console.log(`${search} ${logPrefix}: Tools enabled for this step`)
+		}
+
+		// FIXED: Use correct web search tool definition matching API documentation
+		const tools = shouldUseTools
+			? [
+					{
+						type: 'web_search_20250305', // CHANGED: Use correct tool type from API docs
+						name: 'web_search',
+						max_uses: 3 // ADDED: Limit searches per request
+					}
+				]
+			: undefined
+
+		// ENHANCED: Create API request with conditional tool support
+		const requestParams: any = {
 			model: 'claude-3-7-sonnet-20250219',
-			max_tokens: 4096, // Increased to handle all fields
+			max_tokens: 4096,
 			system: systemPrompt,
 			messages: [{ role: 'user', content: userContent }]
-		})
+		}
+
+		// NEW: Add tools to request if enabled
+		if (tools) {
+			requestParams.tools = tools
+		}
+
+		// FIXED: Implement proper tool execution loop
+		let currentMessages = [...requestParams.messages]
+		let finalText = ''
+		let toolCallCount = 0
+		const maxCalls = Math.min(
+			MAX_TOOL_CALLS_PER_STEP,
+			stepConfigs[stepName as StepName]?.maxToolCalls || MAX_TOOL_CALLS_PER_STEP
+		)
+
+		while (toolCallCount < maxCalls) {
+			// Create request with current message history
+			const currentRequest = {
+				...requestParams,
+				messages: currentMessages
+			}
+
+			const response = await anthropic.messages.create(currentRequest)
+
+			// Log the request ID at debug level
+			console.debug(`${logPrefix} requestId: ${response.id}`)
+
+			// FIXED: Process response content properly
+			let hasToolUse = false
+			let textContent = ''
+
+			for (const content of response.content) {
+				if (content.type === 'text') {
+					textContent += content.text
+				} else if (content.type === 'server_tool_use' && shouldUseTools) {
+					// FIXED: Handle server-side tool use (web search is executed automatically)
+					hasToolUse = true
+					toolCallCount++
+					console.log(`${search} ${logPrefix}: Web search executed - ${content.name}`)
+				} else if (content.type === 'web_search_tool_result') {
+					// FIXED: Handle web search results (automatically provided by API)
+					console.log(`${search} ${logPrefix}: Received web search results`)
+				}
+			}
+
+			// FIXED: Add assistant's response to conversation history
+			currentMessages.push({
+				role: 'assistant',
+				content: response.content
+			})
 
-		// Log the request ID at debug level
-		console.debug(`${logPrefix} requestId: ${response.id}`)
+			// FIXED: Accumulate text content
+			finalText += textContent
 
-		// Ensure the response content is text
-		if (response.content[0].type !== 'text') {
-			throw new Error(`Unexpected content type from API: ${response.content[0].type}`)
+			// FIXED: Break if no tool use or if we've hit limits
+			if (!hasToolUse || toolCallCount >= maxCalls) {
+				break
+			}
+
+			// FIXED: If there was tool use, Claude might continue in the same turn
+			// Check if response has pause_turn stop reason
+			if (response.stop_reason === 'pause_turn') {
+				// Continue the conversation to let Claude finish its turn
+				continue
+			} else {
+				// Tool use complete, break the loop
+				break
+			}
+		}
+
+		// FIXED: Ensure we have text content
+		if (!finalText) {
+			throw new Error('No text content received from Claude')
 		}
 
-		const result = response.content[0].text
 		const elapsed = (Date.now() - startTime) / 1000 // seconds
 
+		// ENHANCED: Log tool usage statistics
+		if (shouldUseTools && toolCallCount > 0) {
+			console.log(`${search} ${logPrefix}: Used ${toolCallCount} web searches`)
+		}
+
 		// Log the full response text at debug level
-		console.debug(`${logPrefix} full response:\n---\n${result}\n---`)
-		return { text: result, durationSec: elapsed }
+		console.debug(`${logPrefix} full response:\n---\n${finalText}\n---`)
+		return { text: finalText, durationSec: elapsed }
+	} catch (error) {
+		// ENHANCED: Better error handling for tool-related failures
+		if (toolsEnabled && (error.message?.includes('tool') || error.message?.includes('search'))) {
+			console.warn(
+				`${search} ${logPrefix}: Tool error, falling back to text-only mode:`,
+				error.message
+			)
+			// Retry without tools on tool-related errors
+			return callClaude(stepName, promptNames, userContent, false)
+		}
+		throw error // Re-throw non-tool errors
 	} finally {
 		console.timeEnd(`${logPrefix}`)
 	}
 }
+// NEW: Function to get step description with tool awareness
+function getStepDescription(stepName: StepName): string {
+	const stepConfig = stepConfigs[stepName]
+	const toolsWillBeUsed = stepConfig?.toolsEnabled && ENABLE_WEB_SEARCH && IS_API_AVAILABLE
+
+	// Return enhanced description if tools are enabled and available
+	if (toolsWillBeUsed && stepConfig?.description) {
+		return stepConfig.description
+	}
 
-// Define user-friendly step descriptions
-const stepDescriptions: Record<StepName, string> = {
-	research: 'Auto-fill missing user inputs',
-	firstDraft: 'Create initial draft',
-	firstCut: 'Remove unnecessary content',
-	firstEdit: 'Improve text flow',
-	toneEdit: 'Adjust tone and style',
-	finalEdit: 'Final polish'
+	// Fallback to standard descriptions
+	const stepDescriptions: Record<StepName, string> = {
+		findTarget: 'Find possible targets',
+		webSearch: 'Research the target',
+		research: 'Auto-fill missing user inputs',
+		firstDraft: 'Create initial draft',
+		firstCut: 'Remove unnecessary content',
+		firstEdit: 'Improve text flow',
+		toneEdit: 'Adjust tone and style',
+		finalEdit: 'Final polish'
+	}
+
+	return stepDescriptions[stepName]
 }
 
 // Function to generate a progress string from the state
+
 function generateProgressString(state: WriteState): string {
 	const pencil = '✏️'
 	const checkmark = '✓'
+	const search = '🔍' // NEW: Icon for tool-enabled steps
+
+	// Get workflow description
+	const workflowDescription = workflowConfigs[state.workflowType].description
 
 	// Generate the progress string
 	let lis = []
 
-	// Completed steps - use descriptions instead of raw step names
+	// ENHANCED: Completed steps with tool usage indicators
 	for (const step of state.completedSteps) {
+		const stepConfig = stepConfigs[step.name]
+		const usedTools = stepConfig?.toolsEnabled && ENABLE_WEB_SEARCH && IS_API_AVAILABLE
+		const icon = usedTools ? `${search}${checkmark}` : checkmark
+
 		lis.push(
-			`<li class="complete">${stepDescriptions[step.name]} (${step.durationSec.toFixed(1)}s) ${checkmark}</li>`
+			`<li class="complete">${getStepDescription(step.name)} (${step.durationSec.toFixed(1)}s) ${icon}</li>`
 		)
 	}
 
-	// Current step - only add if not already completed and state isn't complete
+	// ENHANCED: Current step with tool usage indicator
 	if (
 		state.currentStep &&
 		state.step !== 'complete' &&
 		!state.completedSteps.some((s) => s.name === state.currentStep)
 	) {
-		lis.push(`<li class="current">${stepDescriptions[state.currentStep]} ${pencil}</li>`)
+		const stepConfig = stepConfigs[state.currentStep]
+		const willUseTools = stepConfig?.toolsEnabled && ENABLE_WEB_SEARCH && IS_API_AVAILABLE
+		const icon = willUseTools ? `${search}${pencil}` : pencil
+
+		lis.push(`<li class="current">${getStepDescription(state.currentStep)} ${icon}</li>`)
 	}
 
-	// Remaining steps - filter out any that are in completed steps or current step
+	// ENHANCED: Remaining steps with tool usage preview
 	const completedAndCurrentSteps = [...state.completedSteps.map((s) => s.name), state.currentStep]
 	const filteredRemainingSteps = state.remainingSteps.filter(
 		(step) => !completedAndCurrentSteps.includes(step)
 	)
+
 	lis = lis.concat(
-		filteredRemainingSteps.map((step) => `<li class="pending">${stepDescriptions[step]}</li>`)
+		filteredRemainingSteps.map((step) => {
+			const stepConfig = stepConfigs[step]
+			const willUseTools = stepConfig?.toolsEnabled && ENABLE_WEB_SEARCH && IS_API_AVAILABLE
+			const description = getStepDescription(step)
+			const indicator = willUseTools ? ` ${search}` : ''
+
+			return `<li class="pending">${description}${indicator}</li>`
+		})
 	)
 
 	const listItems = lis.join('')
@@ -286,31 +566,29 @@ function generateProgressString(state: WriteState): string {
 
 	if (state.nextStep === null) {
 		// Process is complete
-		return `<strong>Done (${totalTime.toFixed(1)}s):</strong><ul>${listItems}</ul>`
+		return `<strong>${workflowDescription} - Done (${totalTime.toFixed(1)}s):</strong><ul>${listItems}</ul>`
 	} else {
-		return `<strong>Progress:</strong><ul>${listItems}</ul>`
+		return `<strong>${workflowDescription} - Progress:</strong><ul>${listItems}</ul>`
 	}
 }
 
 // Initialize a new state for the step-by-step process
 function initializeState(userInput: string): WriteState {
-	const allSteps: StepName[] = [
-		'research',
-		'firstDraft',
-		'firstCut',
-		'firstEdit',
-		'toneEdit',
-		'finalEdit'
-	]
+	const { workflowType, cleanedInput } = parseWorkflowType(userInput)
+	const workflowSteps = workflowConfigs[workflowType].steps
+
+	const firstStep = workflowSteps[0]
+	const remainingSteps = workflowSteps.slice(1)
 
 	return {
 		step: 'start',
-		userInput,
+		workflowType,
+		userInput: cleanedInput,
 		email: '',
 		completedSteps: [],
-		currentStep: 'research', // First step
-		remainingSteps: allSteps.slice(1), // All steps except research (which is current)
-		nextStep: 'research'
+		currentStep: firstStep,
+		remainingSteps,
+		nextStep: firstStep
 	}
 }
 
@@ -336,13 +614,61 @@ const stepHandlers: Record<
 	StepName,
 	(state: WriteState) => Promise<{ text: string; durationSec: number }>
 > = {
+	// ENHANCED: Enable tools for target finding
+	findTarget: async (state) => {
+		System_Prompts['Information'] = state.userInput
+
+		// NEW: Check if tools should be enabled for this step
+		const stepConfig = stepConfigs.findTarget
+		const toolsEnabled = stepConfig?.toolsEnabled && ENABLE_WEB_SEARCH
+
+		const result = await callClaude(
+			'findTarget',
+			['Basic', 'Target', 'Information'],
+			'Hello! Please help me find a person to contact!',
+			toolsEnabled // NEW: Pass tool enablement flag
+		)
+
+		state.information = result.text
+		System_Prompts['Information'] = result.text
+
+		return result
+	},
+
+	// ENHANCED: Enable tools for web search (this step is inherently search-based)
+	webSearch: async (state) => {
+		System_Prompts['Information'] = state.userInput
+
+		// NEW: Check if tools should be enabled for this step
+		const stepConfig = stepConfigs.webSearch
+		const toolsEnabled = stepConfig?.toolsEnabled && ENABLE_WEB_SEARCH
+
+		const result = await callClaude(
+			'webSearch',
+			['Basic', 'webSearch', 'Information', 'Results'],
+			'Hello! Please research this person!',
+			toolsEnabled // NEW: Pass tool enablement flag
+		)
+
+		state.information = result.text
+		System_Prompts['Information'] = System_Prompts['Information'] + '\n\n' + result.text
+
+		return result
+	},
+
+	// ENHANCED: Enable tools for research step
 	research: async (state) => {
 		System_Prompts['Information'] = state.userInput
 
+		// NEW: Check if tools should be enabled for this step
+		const stepConfig = stepConfigs.research
+		const toolsEnabled = stepConfig?.toolsEnabled && ENABLE_WEB_SEARCH
+
 		const result = await callClaude(
 			'research',
 			['Basic', 'Mail', 'Information', 'Research'],
-			"Hello! Please update the list of information by replacing all instances of 'undefined' with something that belongs under their respective header based on the rest of the information provided. Thank you!"
+			"Hello! Please update the list of information by replacing all instances of 'undefined' with something that belongs under their respective header based on the rest of the information provided. Thank you!",
+			toolsEnabled // NEW: Pass tool enablement flag
 		)
 
 		state.information = result.text
@@ -351,11 +677,13 @@ const stepHandlers: Record<
 		return result
 	},
 
+	// UNCHANGED: Text processing steps remain without tools for performance
 	firstDraft: async (state) => {
 		return await callClaude(
 			'firstDraft',
 			['Basic', 'Mail', 'First_Draft', 'Results'],
-			'Hello! Please write an email draft using the following information. \n' + state.information
+			'Hello! Please write an email draft using the following information. \n' + state.userInput
+			// NOTE: No toolsEnabled parameter = defaults to false
 		)
 	},
 
@@ -364,6 +692,7 @@ const stepHandlers: Record<
 			'firstCut',
 			['Basic', 'Mail', 'Information', 'First_Cut', 'Results'],
 			'Hello! Please cut the following email draft. \n \n' + state.email
+			// NOTE: No toolsEnabled parameter = defaults to false
 		)
 	},
 
@@ -372,6 +701,7 @@ const stepHandlers: Record<
 			'firstEdit',
 			['Basic', 'Mail', 'Information', 'First_Edit', 'Results'],
 			'Hello! Please edit the following email draft. \n \n' + state.email
+			// NOTE: No toolsEnabled parameter = defaults to false
 		)
 	},
 
@@ -380,6 +710,7 @@ const stepHandlers: Record<
 			'toneEdit',
 			['Basic', 'Mail', 'Information', 'Tone_Edit', 'Results'],
 			'Hello! Please edit the tone of the following email draft. \n \n' + state.email
+			// NOTE: No toolsEnabled parameter = defaults to false
 		)
 	},
 
@@ -388,10 +719,10 @@ const stepHandlers: Record<
 			'finalEdit',
 			['Basic', 'Mail', 'Information', 'Final_Edit', 'Checklist', 'Results'],
 			'Hello! Please edit the following email draft. \n \n' + state.email
+			// NOTE: No toolsEnabled parameter = defaults to false
 		)
 	}
 }
-
 // Process a specific step
 async function processStep(state: WriteState): Promise<WriteState> {
 	if (!state.nextStep) {
@@ -416,8 +747,8 @@ async function processStep(state: WriteState): Promise<WriteState> {
 
 	const result = await stepHandler(state)
 
-	// Update email content (except for research step which updates information)
-	if (currentStep !== 'research') {
+	// Update email content (except for research-like steps which update information)
+	if (/*!['research', 'findTarget', 'webSearch']*/ !['research'].includes(currentStep)) {
 		state.email = result.text
 	}
 
@@ -428,20 +759,13 @@ async function processStep(state: WriteState): Promise<WriteState> {
 		durationSec: result.durationSec
 	})
 
-	// Set next step
-	const allSteps: StepName[] = [
-		'research',
-		'firstDraft',
-		'firstCut',
-		'firstEdit',
-		'toneEdit',
-		'finalEdit'
-	]
-	const currentIndex = allSteps.indexOf(currentStep)
-
-	if (currentIndex !== -1 && currentIndex < allSteps.length - 1) {
-		state.nextStep = allSteps[currentIndex + 1]
-		state.currentStep = allSteps[currentIndex + 1]
+	// Set next step based on workflow configuration
+	const workflowSteps = workflowConfigs[state.workflowType].steps
+	const currentIndex = workflowSteps.indexOf(currentStep)
+
+	if (currentIndex !== -1 && currentIndex < workflowSteps.length - 1) {
+		state.nextStep = workflowSteps[currentIndex + 1]
+		state.currentStep = workflowSteps[currentIndex + 1]
 	} else {
 		// Last step completed, mark as complete
 		state.nextStep = null
@@ -474,7 +798,9 @@ export async function POST({ fetch, request }) {
 			// Continue an existing process
 			try {
 				state = JSON.parse(requestData.stateToken) as WriteState
-				console.log(`${pencil} write: Continuing from step ${state.step}`)
+				console.log(
+					`${pencil} write: Continuing from step ${state.step} (workflow ${state.workflowType})`
+				)
 			} catch (error) {
 				console.error('Error parsing state token:', error)
 				return json({
@@ -497,8 +823,11 @@ export async function POST({ fetch, request }) {
 				} as ChatResponse)
 			}
 
-			// Initialize new state
+			// Initialize new state with workflow parsing
 			state = initializeState(info)
+			console.log(
+				`${pencil} write: Detected workflow type ${state.workflowType}: ${workflowConfigs[state.workflowType].description}`
+			)
 
 			// For initial calls (no stateToken), return progress string without processing
 			if (requestData.stateToken === undefined) {
diff --git a/src/routes/api/write/process-claude-background.ts b/src/routes/api/write/process-claude-background.ts
new file mode 100644
index 000000000..4d633dcf5
--- /dev/null
+++ b/src/routes/api/write/process-claude-background.ts
@@ -0,0 +1,513 @@
+import { env } from '$env/dynamic/private'
+import Anthropic from '@anthropic-ai/sdk'
+
+// Import types and configurations from server
+import type { StepName, WriteState, StepConfig, JobStorage } from './+server'
+import { workflowConfigs } from './+server'
+
+// Environment variables for step processing
+const ANTHROPIC_API_KEY_FOR_WRITE = env.ANTHROPIC_API_KEY_FOR_WRITE || undefined
+const ENABLE_WEB_SEARCH = env.ENABLE_WEB_SEARCH !== 'false'
+const MAX_TOOL_CALLS_PER_STEP = parseInt(env.MAX_TOOL_CALLS_PER_STEP || '3')
+const IS_API_AVAILABLE = !!ANTHROPIC_API_KEY_FOR_WRITE
+
+// Initialize Anthropic client
+const anthropic = IS_API_AVAILABLE
+	? new Anthropic({
+			apiKey: ANTHROPIC_API_KEY_FOR_WRITE
+		})
+	: null
+
+// System prompts for step processing
+const System_Prompts: { [id: string]: string } = {
+	Basic: `You are a helpful AI assistant.
+
+Note: Some fields in the information may begin with a robot emoji (🤖). This indicates the field was automatically generated during research. You can use this information normally, just ignore the emoji marker.`,
+
+	Mail: `
+What follows is the anatomy of a good email, a set of guidelines and
+criteria for writing a good mail. Each paragraph, except the last represents 
+a distinct part of the email.
+
+Subject line:
+The subject line should be short, informative and clearly communicate
+the goal of the mail. It must grab the attention and capture the
+interest of the recipient. Avoid cliché language.
+
+Greeting:
+The greeting must match the tone of the mail. If possible, address the
+recipient by the appropriate title. Keep it short, and mention the reason
+for the mail. Establish a strong connection with the recipient: Are they
+a politician meant to represent you? Is it regarding something they've
+recently done? Make the recipient feel like they owe you an answer.
+
+First paragraph:
+Explain what the purpose of the email is. It must be concise and captivating,
+most people who receive many emails learn to quickly dismiss many. Make
+sure the relation is established and they have a reason to read on. 
+
+Body paragraph:
+The main body of the email should be informative and contain the information
+of the mail. Take great care not to overwhelm the reader: it must be
+logically structured and not too full of facts. The message should remain 
+clear and the relation to the greeting and first paragraph must remain clear.
+It should not be too long, otherwise it might get skimmed. Links to further
+information can be provided.
+
+Conclusion:
+Keep this short and sweet. Make sure it has a CLEAR CALL TO ACTION!
+Restate the reason the recipient should feel the need to act. Thank them
+for their time and/or your ask.
+
+General:
+Make sure the formatting isn't too boring. Write in a manner the recipient
+would respond well to: Do not argue with them, do not mention views they
+probably won't share. Try to play to things they said before and that fit
+their persona. Keep the tone consistent and not too emotional. Do not sound
+crazy.
+`,
+
+	Checklist: `
+Checklist Before Sending
+Message Verification
+    Is the purpose crystal clear?
+    Have I provided necessary context?
+    Is there a specific, achievable call to action?
+    Have I proofread for tone and clarity?
+`,
+
+	First_Draft: `
+Using the information that will be provided by the user, write the mail 
+according to the criteria. Get all the information into the mail. 
+Don't worry about it being too long. Keep the message powerful.
+`,
+
+	First_Cut: `
+You will be provided with an email by the user. 
+Remove redundant information and clean up the structure. The point of this pass is 
+to have the structure clear and the mail slightly longer than needed. The message 
+should be clear, the information still mostly present, with only what is 
+absolutely necessary being removed.
+`,
+
+	First_Edit: `
+You will be provided with an email by the user. The following points are paramount:
+Make sure the flow of information is natural. All paragraphs should be
+connected in a sensical manner. Remove odd, unfitting or overly emotional
+language. Make sure the paragraphs fulfill their roles.
+`,
+
+	Tone_Edit: `
+You will be provided with an email by the user. The following points are paramount:
+Adjust the language to match recipient's communication style. Remove potentially 
+offensive or dismissive language. Ensure the tone matches the relationship and 
+purpose. Make sure the points and information is relevant for the recipient. 
+Assume the recipient's position: How would they react to the mail? What information 
+would resonate with them? What wouldn't? Do not compromise on the message.
+`,
+
+	Final_Edit: `
+You will be provided with an email by the user. Make sure the email matches the 
+criteria initially described. Check spelling, grammar and tone.
+`,
+
+	Research: `
+Please replace all mentions of 'undefined' with the apropriate information that should
+go in that space, derived from the rest of the information. 
+
+Important: For any field you fill in that was originally 'undefined' or empty, prefix 
+your answer with a robot emoji (🤖) to indicate it was automatically generated.
+
+Example:
+Original: "Preferred communication style: undefined"
+Your output: "Preferred communication style: 🤖 Formal but approachable"
+
+Please remember that you are addressing this person, and try to make all inferences based on the information provided and your own knowledge. Err on the side of caution: if you are unsure, be polite and neutral.
+
+Output the full information, including your edits. Output nothing else.
+`,
+
+	Target: `
+Please use your internet search capability to find individuals involved with AI safety who match the following description.
+
+For each person you find (aim for 3-5 people), please provide:
+1. Name and current position
+2. Why they're relevant to AI safety
+3. Their organization
+4. Brief note on their public stance on AI safety
+
+Please cite your sources for each person.
+`,
+
+	webSearch: `
+Please use your internet search capability to research [Person's Name] who is [current role] at [organization/affiliation]. I plan to contact them about AI safety concerns.
+
+Search for and provide:
+1. Professional background (education, career history, notable positions)
+2. Their involvement with AI issues (policy positions, public statements, initiatives, articles, interviews)
+3. Their public views on AI development and safety (with direct quotes where possible)
+4. Recent activities related to technology policy or AI (last 6-12 months)
+5. Communication style and key terms they use when discussing technology issues
+6. Notable connections (organizations, committees, coalitions, or influential individuals they work with)
+7. Contact information (professional email or official channels if publicly available)
+
+Please cite all sources you use and only include information you can verify through your internet search. If you encounter conflicting information, note this and provide the most reliable source.
+
+BE BRIEF! This is extremely important. Try to output only a few lines of text for each questions.
+`,
+
+	Results: `
+Only reply with the final results, a.k.a. the final email, and absolutely nothing else.
+`
+}
+
+// Step configurations
+export const stepConfigs: Record<StepName, StepConfig> = {
+	findTarget: {
+		toolsEnabled: true,
+		maxToolCalls: 5,
+		description: 'Find possible targets (using web search)'
+	},
+	webSearch: {
+		toolsEnabled: true,
+		maxToolCalls: 3,
+		description: 'Research the target (using web search)'
+	},
+	research: {
+		toolsEnabled: false,
+		maxToolCalls: 2,
+		description: 'Auto-fill missing user inputs'
+	},
+	firstDraft: { toolsEnabled: false },
+	firstCut: { toolsEnabled: false },
+	firstEdit: { toolsEnabled: false },
+	toneEdit: { toolsEnabled: false },
+	finalEdit: { toolsEnabled: false }
+}
+
+// Enhanced callClaude function with tool support
+export async function callClaude(
+	stepName: string,
+	promptNames: string[],
+	userContent: string,
+	toolsEnabled: boolean = false
+): Promise<{ text: string; durationSec: number }> {
+	const pencil = '✏️'
+	const search = '🔍'
+	const logPrefix = `${pencil} write:${stepName}`
+	const startTime = Date.now()
+
+	console.time(`${logPrefix}`)
+
+	try {
+		// Check if the API client is available
+		if (!anthropic) {
+			throw new Error('Anthropic API client is not initialized. API key is missing.')
+		}
+
+		// Combine all the specified prompts
+		const systemPrompt = promptNames.map((name) => System_Prompts[name]).join('')
+
+		// Determine if tools should be included in this call
+		const shouldUseTools = toolsEnabled && ENABLE_WEB_SEARCH && IS_API_AVAILABLE
+
+		// Log tool usage status
+		if (shouldUseTools) {
+			console.log(`${search} ${logPrefix}: Tools enabled for this step`)
+		}
+
+		// Use correct web search tool definition matching API documentation
+		const tools = shouldUseTools
+			? [
+					{
+						type: 'web_search_20250305',
+						name: 'web_search',
+						max_uses: 3
+					}
+				]
+			: undefined
+
+		// Create API request with conditional tool support
+		const requestParams: any = {
+			model: 'claude-3-7-sonnet-20250219',
+			max_tokens: 4096,
+			system: systemPrompt,
+			messages: [{ role: 'user', content: userContent }]
+		}
+
+		// Add tools to request if enabled
+		if (tools) {
+			requestParams.tools = tools
+		}
+
+		// Implement proper tool execution loop
+		let currentMessages = [...requestParams.messages]
+		let finalText = ''
+		let toolCallCount = 0
+		const maxCalls = Math.min(
+			MAX_TOOL_CALLS_PER_STEP,
+			stepConfigs[stepName as StepName]?.maxToolCalls || MAX_TOOL_CALLS_PER_STEP
+		)
+
+		while (toolCallCount < maxCalls) {
+			// Create request with current message history
+			const currentRequest = {
+				...requestParams,
+				messages: currentMessages
+			}
+
+			const response = await anthropic.messages.create(currentRequest)
+
+			// Log the request ID at debug level
+			console.debug(`${logPrefix} requestId: ${response.id}`)
+
+			// Process response content properly
+			let hasToolUse = false
+			let textContent = ''
+
+			for (const content of response.content) {
+				if (content.type === 'text') {
+					textContent += content.text
+				} else if (content.type === 'server_tool_use' && shouldUseTools) {
+					// Handle server-side tool use (web search is executed automatically)
+					hasToolUse = true
+					toolCallCount++
+					console.log(`${search} ${logPrefix}: Web search executed - ${content.name}`)
+				} else if (content.type === 'web_search_tool_result') {
+					// Handle web search results (automatically provided by API)
+					console.log(`${search} ${logPrefix}: Received web search results`)
+				}
+			}
+
+			// Add assistant's response to conversation history
+			currentMessages.push({
+				role: 'assistant',
+				content: response.content
+			})
+
+			// Accumulate text content
+			finalText += textContent
+
+			// Break if no tool use or if we've hit limits
+			if (!hasToolUse || toolCallCount >= maxCalls) {
+				break
+			}
+
+			// If there was tool use, Claude might continue in the same turn
+			// Check if response has pause_turn stop reason
+			if (response.stop_reason === 'pause_turn') {
+				// Continue the conversation to let Claude finish its turn
+				continue
+			} else {
+				// Tool use complete, break the loop
+				break
+			}
+		}
+
+		// Ensure we have text content
+		if (!finalText) {
+			throw new Error('No text content received from Claude')
+		}
+
+		const elapsed = (Date.now() - startTime) / 1000 // seconds
+
+		// Log tool usage statistics
+		if (shouldUseTools && toolCallCount > 0) {
+			console.log(`${search} ${logPrefix}: Used ${toolCallCount} web searches`)
+		}
+
+		// Log the full response text at debug level
+		console.debug(`${logPrefix} full response:\n---\n${finalText}\n---`)
+		return { text: finalText, durationSec: elapsed }
+	} catch (error) {
+		// Better error handling for tool-related failures
+		if (toolsEnabled && (error.message?.includes('tool') || error.message?.includes('search'))) {
+			console.warn(
+				`${search} ${logPrefix}: Tool error, falling back to text-only mode:`,
+				error.message
+			)
+			// Retry without tools on tool-related errors
+			return callClaude(stepName, promptNames, userContent, false)
+		}
+		throw error // Re-throw non-tool errors
+	} finally {
+		console.timeEnd(`${logPrefix}`)
+	}
+}
+
+// Define step handlers in a map for easy lookup
+const stepHandlers: Record<
+	StepName,
+	(state: WriteState) => Promise<{ text: string; durationSec: number }>
+> = {
+	// Enable tools for target finding
+	findTarget: async (state) => {
+		System_Prompts['Information'] = state.userInput
+
+		// Check if tools should be enabled for this step
+		const stepConfig = stepConfigs.findTarget
+		const toolsEnabled = stepConfig?.toolsEnabled && ENABLE_WEB_SEARCH
+
+		const result = await callClaude(
+			'findTarget',
+			['Basic', 'Target', 'Information'],
+			'Hello! Please help me find a person to contact!',
+			toolsEnabled
+		)
+
+		state.information = result.text
+		System_Prompts['Information'] = result.text
+
+		return result
+	},
+
+	// Enable tools for web search
+	webSearch: async (state) => {
+		System_Prompts['Information'] = state.userInput
+
+		// Check if tools should be enabled for this step
+		const stepConfig = stepConfigs.webSearch
+		const toolsEnabled = stepConfig?.toolsEnabled && ENABLE_WEB_SEARCH
+
+		const result = await callClaude(
+			'webSearch',
+			['Basic', 'webSearch', 'Information', 'Results'],
+			'Hello! Please research this person!',
+			toolsEnabled
+		)
+
+		state.information = result.text
+		System_Prompts['Information'] = System_Prompts['Information'] + '\n\n' + result.text
+
+		return result
+	},
+
+	// Enable tools for research step
+	research: async (state) => {
+		System_Prompts['Information'] = state.userInput
+
+		// Check if tools should be enabled for this step
+		const stepConfig = stepConfigs.research
+		const toolsEnabled = stepConfig?.toolsEnabled && ENABLE_WEB_SEARCH
+
+		const result = await callClaude(
+			'research',
+			['Basic', 'Mail', 'Information', 'Research'],
+			"Hello! Please update the list of information by replacing all instances of 'undefined' with something that belongs under their respective header based on the rest of the information provided. Thank you!",
+			toolsEnabled
+		)
+
+		state.information = result.text
+		System_Prompts['Information'] = result.text
+
+		return result
+	},
+
+	// Text processing steps remain without tools for performance
+	firstDraft: async (state) => {
+		return await callClaude(
+			'firstDraft',
+			['Basic', 'Mail', 'First_Draft', 'Results'],
+			'Hello! Please write an email draft using the following information. \n' + state.userInput
+		)
+	},
+
+	firstCut: async (state) => {
+		return await callClaude(
+			'firstCut',
+			['Basic', 'Mail', 'Information', 'First_Cut', 'Results'],
+			'Hello! Please cut the following email draft. \n \n' + state.email
+		)
+	},
+
+	firstEdit: async (state) => {
+		return await callClaude(
+			'firstEdit',
+			['Basic', 'Mail', 'Information', 'First_Edit', 'Results'],
+			'Hello! Please edit the following email draft. \n \n' + state.email
+		)
+	},
+
+	toneEdit: async (state) => {
+		return await callClaude(
+			'toneEdit',
+			['Basic', 'Mail', 'Information', 'Tone_Edit', 'Results'],
+			'Hello! Please edit the tone of the following email draft. \n \n' + state.email
+		)
+	},
+
+	finalEdit: async (state) => {
+		return await callClaude(
+			'finalEdit',
+			['Basic', 'Mail', 'Information', 'Final_Edit', 'Checklist', 'Results'],
+			'Hello! Please edit the following email draft. \n \n' + state.email
+		)
+	}
+}
+
+// Process a specific step
+export async function processStep(state: WriteState): Promise<WriteState> {
+	if (!state.nextStep) {
+		return {
+			...state,
+			step: 'complete',
+			currentStep: null
+		}
+	}
+
+	const currentStep = state.nextStep as StepName
+	state.currentStep = currentStep
+
+	// Update remaining steps - remove current step from remaining
+	state.remainingSteps = state.remainingSteps.filter((step) => step !== currentStep)
+
+	// Execute the step using the step handler from the map
+	const stepHandler = stepHandlers[currentStep]
+	if (!stepHandler) {
+		throw new Error(`Unknown step: ${currentStep}`)
+	}
+
+	const result = await stepHandler(state)
+
+	// Update email content (except for research-like steps which update information)
+	if (!['research'].includes(currentStep)) {
+		state.email = result.text
+	}
+
+	// Update state with results - current step is now completed
+	state.step = currentStep
+	state.completedSteps.push({
+		name: currentStep,
+		durationSec: result.durationSec
+	})
+
+	// Set next step based on workflow configuration
+	// Note: This requires access to workflowConfigs which should be imported from the server file
+	const workflowSteps = getWorkflowSteps(state.workflowType)
+	const currentIndex = workflowSteps.indexOf(currentStep)
+
+	if (currentIndex !== -1 && currentIndex < workflowSteps.length - 1) {
+		state.nextStep = workflowSteps[currentIndex + 1]
+		state.currentStep = workflowSteps[currentIndex + 1]
+	} else {
+		// Last step completed, mark as complete
+		state.nextStep = null
+		state.currentStep = null
+		state.step = 'complete'
+	}
+
+	return state
+}
+
+// Helper function to get workflow steps (this will need to be imported from server)
+function getWorkflowSteps(workflowType: string): StepName[] {
+	// This should be imported from the server file or passed as a parameter
+	// For now, returning a placeholder
+	const workflowConfigs = {
+		'1': ['findTarget'],
+		'2': ['webSearch', 'research'],
+		'3': ['research'],
+		'4': ['firstDraft', 'firstCut', 'firstEdit', 'toneEdit', 'finalEdit']
+	}
+	return workflowConfigs[workflowType] || []
+}
diff --git a/src/routes/write/+page.svelte b/src/routes/write/+page.svelte
index a864d3f5c..f42bb0b57 100644
--- a/src/routes/write/+page.svelte
+++ b/src/routes/write/+page.svelte
@@ -23,18 +23,28 @@
 
 	// Use a unique localStorage key to avoid conflicts with other pages
 	const STORAGE_KEY = 'email_writer_messages'
+	// CLAUDE CHANGE: Added storage key for form data
+	const FORM_DATA_STORAGE_KEY = 'email_writer_form_data'
 
 	let messages: Message[] =
 		typeof localStorage !== 'undefined' ? JSON.parse(localStorage.getItem(STORAGE_KEY) || '[]') : []
 
 	// Array for form
-	let input_arr = new Array<string>(35)
 	let loading = false
 	let apiAvailable = true // Default to true, will be updated after first API call
 	const maxMessages = 20
 
 	// Organizing the form questions into sections and subsections
-	const formSections: FieldSection[] = [
+	const formSections_Target: FieldSection[] = [
+		{
+			title: 'Researching a person',
+			subsections: [
+				{
+					title: 'To research a person, fill out the following fields.',
+					questions: ["Person's Name", 'Current Role', 'Organization/Affiliation']
+				}
+			]
+		},
 		{
 			title: 'Personal Context',
 			subsections: [
@@ -92,27 +102,47 @@
 					]
 				}
 			]
-		},
+		}
+	]
+
+	// Flatten the questions array for accessing by index
+	const paragraphText_Target: string[] = []
+	formSections_Target.forEach((section) => {
+		section.subsections.forEach((subsection) => {
+			subsection.questions.forEach((question) => {
+				paragraphText_Target.push(question)
+			})
+		})
+	})
+
+	const formSections_Research: FieldSection[] = [
 		{
-			title: 'Information Needed About the Message',
+			title: 'Finding a target',
 			subsections: [
 				{
-					title: 'Content Requirements',
-					questions: ['Specific outcome desired', 'Concrete action requested'] /*
-					questions: [
-						'Clear, singular objective',
-						'Specific outcome desired',
-						'Concrete action requested'
-					]*/
-				},
-				{
-					title: 'Supporting Evidence',
+					title: "Specify what sort of target you're looking for, and where.",
 					questions: [
-						'Relevant facts',
-						'Context for the request',
-						'Potential impact or consequences'
+						'If you have certain institutions in mind, mention those. Otherwise, your local representative could be a good place to start.'
 					]
-				},
+				}
+			]
+		}
+	]
+
+	// Flatten the questions array for accessing by index
+	const paragraphText_Research: string[] = []
+	formSections_Research.forEach((section) => {
+		section.subsections.forEach((subsection) => {
+			subsection.questions.forEach((question) => {
+				paragraphText_Research.push(question)
+			})
+		})
+	})
+
+	const formSections_MessageDetails: FieldSection[] = [
+		{
+			title: 'Message Details',
+			subsections: [
 				{
 					title: 'Logical Structure',
 					questions: [
@@ -164,25 +194,109 @@
 		}
 	]
 
+	// CLAUDE CHANGE: Fixed the population of paragraphText_MessageDetails - was incorrectly using formSections_Research
+	const paragraphText_MessageDetails: string[] = []
+	formSections_MessageDetails.forEach((section) => {
+		section.subsections.forEach((subsection) => {
+			subsection.questions.forEach((question) => {
+				paragraphText_MessageDetails.push(question)
+			})
+		})
+	})
+
+	const formSections_Message: FieldSection[] = [
+		{
+			title: 'What is your Message?',
+			subsections: [
+				{
+					title: 'Content Requirements',
+					questions: ['Specific outcome desired', 'Concrete action requested'] /*
+					questions: [
+						'Clear, singular objective',
+						'Specific outcome desired',
+						'Concrete action requested'
+					]*/
+				},
+				{
+					title: 'Supporting Evidence',
+					questions: [
+						'Relevant facts',
+						'Context for the request',
+						'Potential impact or consequences'
+					]
+				}
+			]
+		}
+	]
+
 	// Flatten the questions array for accessing by index
-	const paragraphText: string[] = []
-	formSections.forEach((section) => {
+	const paragraphText_Message: string[] = []
+	formSections_Message.forEach((section) => {
 		section.subsections.forEach((subsection) => {
 			subsection.questions.forEach((question) => {
-				paragraphText.push(question)
+				paragraphText_Message.push(question)
 			})
 		})
 	})
 
+	// CLAUDE CHANGE: Added mapping functions to get correct arrays based on active form
+	function getCurrentInputArray(): string[] {
+		switch (activeForm) {
+			case 'form1':
+				return form1_input_arr
+			case 'form2':
+				return form2_input_arr
+			case 'form3':
+				return form3_input_arr
+			case 'form4':
+				return form4_input_arr
+			case 'form5':
+				return form2_input_arr.concat(form3_input_arr, form4_input_arr)
+			default:
+				return form2_input_arr
+		}
+	}
+
+	function getCurrentQuestionArray(): string[] {
+		switch (activeForm) {
+			case 'form1':
+				return paragraphText_Research
+			case 'form2':
+				return paragraphText_Target
+			case 'form3':
+				return paragraphText_Message
+			case 'form4':
+				return paragraphText_MessageDetails
+			case 'form5':
+				let allText = paragraphText_Target.concat(
+					paragraphText_Message,
+					paragraphText_MessageDetails
+				)
+				return allText
+			default:
+				return paragraphText_Target
+		}
+	}
+
 	function clear_arr(arr: string[]) {
 		for (var i in arr) {
 			arr[i] = ''
 		}
 	}
 
+	// CLAUDE CHANGE: Updated clear function to clear current form and reset to form1
 	function clear() {
 		messages = []
 		localStorage.setItem(STORAGE_KEY, JSON.stringify(messages))
+
+		// Clear all form arrays
+		clear_arr(form1_input_arr)
+		clear_arr(form2_input_arr)
+		clear_arr(form3_input_arr)
+		clear_arr(form4_input_arr)
+
+		// Clear form data from localStorage
+		localStorage.removeItem(FORM_DATA_STORAGE_KEY)
 	}
 
 	function copy() {
@@ -194,39 +308,111 @@
 		)
 	}
 
-	// Magic test function that populates three specific fields and starts the process - for demo purposes
+	// CLAUDE CHANGE: Updated runTest to work with currently active form and provide appropriate test data
 	function runTest() {
 		// Clear any existing chat
 		clear()
 
-		// Clear all input fields
-		clear_arr(input_arr)
+		// Get current form arrays
+		const currentInputArray = getCurrentInputArray()
+		const currentQuestionArray = getCurrentQuestionArray()
 
-		// Find index for specific fields based on their question text
-		const roleAuthorityIndex = paragraphText.findIndex(
-			(q) => q === 'Understanding their role and potential authority'
-		)
-		const objectiveIndex = paragraphText.findIndex((q) => q === 'Concrete action requested')
-		const outcomeIndex = paragraphText.findIndex((q) => q === 'Specific outcome desired')
+		// Clear current input fields
+		clear_arr(currentInputArray)
 
-		input_arr[roleAuthorityIndex] = 'You are writing for a child of about thirteen years old.'
-		input_arr[objectiveIndex] =
-			'Child understands that the current fast worldwide race to build AGI before we know how to make it safe could mean everybody dies. It sounds like science fiction, but experts are increasingly concerned. Of course the grown-ups should do something about this, but because it can be hard for them to admit to themselves that those they love could be in so much danger, then even once they hear about the problem they find reasons not to believe or act. Being sincere, kind and persistent with them that you need protecting can help everybody in the world.'
-		input_arr[outcomeIndex] =
-			"The child isn't too scared, but thinks this is a very important problem all their friends, family and teachers should know about and be actively working to solve."
+		// Provide test data based on active form
+		switch (activeForm) {
+			case 'form1':
+				// Test data for research form
+				if (currentInputArray.length > 0) {
+					currentInputArray[0] =
+						'Local city council member or school board representative who handles education policy and child safety issues. Location: California, Los Angeles.'
+				}
+				break
+
+			case 'form2':
+				// Test data for target/personal context form
+				const roleAuthorityIndex = currentQuestionArray.findIndex(
+					(q) => q === 'Understanding their role and potential authority'
+				)
+				if (roleAuthorityIndex >= 0) {
+					currentInputArray[roleAuthorityIndex] =
+						'You are writing for a local government official with decision-making authority over education and safety policies.'
+				}
+				break
+
+			case 'form3':
+				// Test data for message form
+				const objectiveIndex = currentQuestionArray.findIndex(
+					(q) => q === 'Concrete action requested'
+				)
+				const outcomeIndex = currentQuestionArray.findIndex((q) => q === 'Specific outcome desired')
+
+				if (objectiveIndex >= 0) {
+					currentInputArray[objectiveIndex] =
+						'Official takes action to ensure AI safety education and policies are implemented in local institutions.'
+				}
+				if (outcomeIndex >= 0) {
+					currentInputArray[outcomeIndex] =
+						'Local community becomes informed about AI risks and appropriate safety measures are put in place.'
+				}
+				break
+
+			case 'form4':
+				// Test data for message details form
+				const urgencyIndex = currentQuestionArray.findIndex((q) => q === 'Urgency of the request')
+				const toneIndex = currentQuestionArray.findIndex(
+					(q) => q === 'Balancing professionalism and approachability'
+				)
+
+				if (urgencyIndex >= 0) {
+					currentInputArray[urgencyIndex] =
+						'High urgency due to rapidly advancing AI development timeline.'
+				}
+				if (toneIndex >= 0) {
+					currentInputArray[toneIndex] =
+						'Professional but accessible tone that conveys seriousness without being alarmist.'
+				}
+				break
+		}
 
 		sendMessage()
 	}
 
+	// CLAUDE CHANGE: Updated sendMessage to work with currently active form only
 	async function sendMessage() {
+		const currentInputArray = getCurrentInputArray()
+		const currentQuestionArray = getCurrentQuestionArray()
+
 		let input = ''
-		for (var i in paragraphText) {
-			input = input + paragraphText[i] + ':\n' + input_arr[i] + '\n\n'
+		switch (activeForm) {
+			case 'form1':
+				input = input + '[1]'
+				break
+
+			case 'form2':
+				input = input + '[2]'
+				break
+
+			case 'form4':
+				input = input + '[3]'
+				break
+
+			case 'form5':
+				input = input + '[4]'
+				break
 		}
+		for (let i = 0; i < currentQuestionArray.length; i++) {
+			input =
+				input + currentQuestionArray[i] + ':\n' + (currentInputArray[i] || 'undefined') + '\n\n'
+		}
+
 		messages = [...messages, { content: input, role: 'user' }]
 
-		clear_arr(input_arr)
+		// Clear current form fields
+		clear_arr(currentInputArray)
 		loading = true
+		console.log(input)
 
 		try {
 			// First request - get initial progress message (no stateToken)
@@ -294,6 +480,8 @@
 
 			// Process the response
 			const data = await response.json()
+			console.log('DATA: \n' + data)
+			console.log(data.information)
 
 			// Update API availability
 			apiAvailable = data.apiAvailable !== false
@@ -332,23 +520,31 @@
 				}
 			}
 
-			// Update form fields if we have information from the research step
+			// CLAUDE CHANGE: Updated auto-fill logic to work with currently active form
 			if (data.information) {
+				const currentInputArray = getCurrentInputArray()
+				const currentQuestionArray = getCurrentQuestionArray()
+
 				// Parse the information string to extract field values
 				const lines = data.information.split('\n')
 				let currentField = -1
 
 				for (let line of lines) {
-					// Look for field headers matching our paragraphText array
-					const fieldIndex = paragraphText.findIndex((text) => line.trim().startsWith(text + ':'))
+					// Look for field headers matching our current question array
+					const fieldIndex = currentQuestionArray.findIndex((text) =>
+						line.trim().startsWith(text + ':')
+					)
 
 					if (fieldIndex >= 0) {
 						currentField = fieldIndex
 					} else if (currentField >= 0 && line.trim()) {
 						// Only update when there's actual content and the field is empty
 						const lineContent = line.trim()
-						if (lineContent && (!input_arr[currentField] || input_arr[currentField] === '')) {
-							input_arr[currentField] = lineContent
+						if (
+							lineContent &&
+							(!currentInputArray[currentField] || currentInputArray[currentField] === '')
+						) {
+							currentInputArray[currentField] = lineContent
 						}
 					}
 				}
@@ -356,6 +552,8 @@
 
 			// Save messages to localStorage
 			localStorage.setItem(STORAGE_KEY, JSON.stringify(messages))
+			// CLAUDE CHANGE: Also save form data
+			saveFormData()
 
 			// If not complete, continue with the next step
 			if (!data.complete && data.stateToken) {
@@ -375,7 +573,35 @@
 		}
 	}
 
+	// CLAUDE CHANGE: Added form data persistence functions
+	function saveFormData() {
+		const formData = {
+			form1_input_arr,
+			form2_input_arr,
+			form3_input_arr,
+			form4_input_arr,
+			activeForm
+		}
+		localStorage.setItem(FORM_DATA_STORAGE_KEY, JSON.stringify(formData))
+	}
+
+	function loadFormData() {
+		const saved = localStorage.getItem(FORM_DATA_STORAGE_KEY)
+		if (saved) {
+			const formData = JSON.parse(saved)
+			form1_input_arr = formData.form1_input_arr || new Array<string>(paragraphText_Research.length)
+			form2_input_arr = formData.form2_input_arr || new Array<string>(paragraphText_Target.length)
+			form3_input_arr = formData.form3_input_arr || new Array<string>(paragraphText_Message.length)
+			form4_input_arr =
+				formData.form4_input_arr || new Array<string>(paragraphText_MessageDetails.length)
+			activeForm = formData.activeForm || 'form1'
+		}
+	}
+
 	onMount(async () => {
+		// CLAUDE CHANGE: Load form data on mount
+		loadFormData()
+
 		// Check API availability on component mount
 		try {
 			const response = await fetch('api/write')
@@ -397,11 +623,33 @@
 		*/
 	}
 
-	// Function to get the index of a question across all sections
-	function getQuestionIndex(question: string): number {
-		return paragraphText.findIndex((text) => text === question)
+	// FORM FUNCTIONS //
+
+	// Add these variables for form toggling
+	let activeForm = 'form1' // Default active form
+
+	// CLAUDE CHANGE: Updated setActiveForm to save form data when switching
+	function setActiveForm(formId: string) {
+		saveFormData() // Save current state before switching
+		activeForm = formId
+		saveFormData() // Save current state after switching to save activeForm
+		console.log(formId)
 	}
 
+	function writeMail() {
+		setActiveForm('form5')
+		sendMessage()
+	}
+
+	// UNTESTED AND GENERATED BY AI
+	// Create separate arrays for each form's inputs
+	//let input_arr = Array(formSections_Target.flatMap(s => s.subsections.flatMap(ss => ss.questions)).length).fill('');
+	// CLAUDE CHANGE: Fixed form array initialization to use correct lengths
+	let form1_input_arr = Array(paragraphText_Research.length).fill('')
+	let form2_input_arr = Array(paragraphText_Target.length).fill('')
+	let form3_input_arr = Array(paragraphText_Message.length).fill('')
+	let form4_input_arr = Array(paragraphText_MessageDetails.length).fill('')
+
 	// Top of the page
 	const title = `Write Email Content`
 	const description = `This (beta!) webpage lets you write email content (with LLM assistance.)`
@@ -453,13 +701,41 @@
 			particular hardcoded values, and starts writing content.
 		</p>
 
+		<!-- Form toggle bar -->
+		<div class="control-buttons">
+			<button
+				class="button {activeForm === 'form1' ? 'active' : ''}"
+				on:click={() => setActiveForm('form1')}
+			>
+				Finding A Target
+			</button>
+			<button
+				class="button {activeForm === 'form2' ? 'active' : ''}"
+				on:click={() => setActiveForm('form2')}
+			>
+				Personal Context
+			</button>
+			<button
+				class="button {activeForm === 'form3' ? 'active' : ''}"
+				on:click={() => setActiveForm('form3')}
+			>
+				The Message
+			</button>
+			<button
+				class="button {activeForm === 'form4' ? 'active' : ''}"
+				on:click={() => setActiveForm('form4')}
+			>
+				Message details
+			</button>
+		</div>
+
 		<div class="control-buttons">
 			<button
 				on:click={sendMessage}
-				disabled={!apiAvailable || loading}
+				disabled={!apiAvailable || loading || activeForm === 'form3'}
 				class="button {!apiAvailable ? 'button--disabled' : ''}"
 			>
-				Write Content
+				Autofill
 			</button>
 			<button on:click={runTest} class="button" disabled={!apiAvailable || loading}>
 				(Demo for beta)
@@ -470,6 +746,13 @@
 			<button on:click={clear} class="button" disabled={loading || messages.length === 0}>
 				Reset All
 			</button>
+			<button
+				on:click={writeMail}
+				disabled={!apiAvailable || loading}
+				class="button {!apiAvailable ? 'button--disabled' : ''}"
+			>
+				Write Mail
+			</button>
 		</div>
 	</div>
 
@@ -496,40 +779,117 @@
 		<button class="button" on:click={copy}>Copy Content</button>
 		<button class="button" on:click={clear}>Reset All</button>
 	{:else}
-		<form on:submit|preventDefault>
-			<!-- Render form sections using the structured data -->
-			{#each formSections as section, sectionIndex}
-				<h1>{section.title}</h1>
-
-				{#each section.subsections as subsection, subsectionIndex}
-					<h2>{subsection.title}</h2>
-
-					<!-- Special handling for 'Content Requirements' to add an h3 for 'Precise Purpose' -->
-					{#if subsection.title === 'Content Requirements'}
-						<h3>Precise Purpose</h3>
-					{/if}
-
-					{#each subsection.questions as question, questionIndex}
-						<!-- Calculate the global index for this question -->
-						{@const globalIndex = getQuestionIndex(question)}
-
-						<!-- Add special h3 headers for specific subsections -->
-						{#if subsection.title === 'Supporting Evidence' && questionIndex === 0}
-							<h3>Supporting Evidence</h3>
-						{:else if subsection.title === 'Logical Structure' && questionIndex === 0}
-							<h3>Logical Structure</h3>
-						{/if}
-
-						<p>{question}</p>
-						<textarea
-							placeholder="Type here (Question {globalIndex + 1})"
-							bind:value={input_arr[globalIndex]}
-							on:keydown={handleKeyDown}
-						></textarea>
+		<!-- Form container with conditional display based on active form -->
+		<!-- CLAUDE MODIFICATION: Modified the form container section to show forms -->
+		<div class="form-container">
+			<!-- Form 1 - Remains unchanged -->
+			{#if activeForm === 'form1'}
+				<form on:submit|preventDefault>
+					{#each formSections_Research as section, sectionIndex}
+						<h1>{section.title}</h1>
+						{#each section.subsections as subsection, subsectionIndex}
+							<h2>{subsection.title}</h2>
+
+							{#each subsection.questions as question, questionIndex}
+								{@const globalIndex = paragraphText_Research.findIndex((text) => text === question)}
+
+								<p>{question}</p>
+								<textarea
+									placeholder="Type here (Question {globalIndex + 1})"
+									bind:value={form1_input_arr[globalIndex]}
+									on:keydown={handleKeyDown}
+								></textarea>
+							{/each}
+						{/each}
+					{/each}
+				</form>
+			{/if}
+
+			<!-- Form 2 - Now includes merged PersonResearch and Target sections -->
+			{#if activeForm === 'form2'}
+				<form on:submit|preventDefault>
+					<!-- Render form sections using the structured data -->
+					{#each formSections_Target as section, sectionIndex}
+						<h1>{section.title}</h1>
+						{#each section.subsections as subsection, subsectionIndex}
+							<h2>{subsection.title}</h2>
+
+							<!-- Special handling for 'Content Requirements' to add an h3 for 'Precise Purpose' -->
+							{#if subsection.title === 'Content Requirements'}
+								<h3>Precise Purpose</h3>
+							{/if}
+
+							{#each subsection.questions as question, questionIndex}
+								<!-- Calculate the global index for this question -->
+								{@const globalIndex = paragraphText_Target.findIndex((text) => text === question)}
+
+								<!-- Add special h3 headers for specific subsections -->
+								{#if subsection.title === 'Supporting Evidence' && questionIndex === 0}
+									<h3>Supporting Evidence</h3>
+								{:else if subsection.title === 'Logical Structure' && questionIndex === 0}
+									<h3>Logical Structure</h3>
+								{/if}
+
+								<p>{question}</p>
+								<textarea
+									placeholder="Type here (Question {globalIndex + 1})"
+									bind:value={form2_input_arr[globalIndex]}
+									on:keydown={handleKeyDown}
+								></textarea>
+							{/each}
+						{/each}
 					{/each}
-				{/each}
-			{/each}
-		</form>
+				</form>
+			{/if}
+
+			<!-- Form 3 -->
+			{#if activeForm === 'form3'}
+				<form on:submit|preventDefault>
+					{#each formSections_Message as section, sectionIndex}
+						<h1>{section.title}</h1>
+						{#each section.subsections as subsection, subsectionIndex}
+							<h2>{subsection.title}</h2>
+
+							{#each subsection.questions as question, questionIndex}
+								{@const globalIndex = paragraphText_Message.findIndex((text) => text === question)}
+
+								<p>{question}</p>
+								<textarea
+									placeholder="Type here (Question {globalIndex + 1})"
+									bind:value={form3_input_arr[globalIndex]}
+									on:keydown={handleKeyDown}
+								></textarea>
+							{/each}
+						{/each}
+					{/each}
+				</form>
+			{/if}
+
+			<!-- Form 4 -->
+			{#if activeForm === 'form4'}
+				<form on:submit|preventDefault>
+					{#each formSections_MessageDetails as section, sectionIndex}
+						<h1>{section.title}</h1>
+						{#each section.subsections as subsection, subsectionIndex}
+							<h2>{subsection.title}</h2>
+
+							{#each subsection.questions as question, questionIndex}
+								{@const globalIndex = paragraphText_MessageDetails.findIndex(
+									(text) => text === question
+								)}
+
+								<p>{question}</p>
+								<textarea
+									placeholder="Type here (Question {globalIndex + 1})"
+									bind:value={form4_input_arr[globalIndex]}
+									on:keydown={handleKeyDown}
+								></textarea>
+							{/each}
+						{/each}
+					{/each}
+				</form>
+			{/if}
+		</div>
 	{/if}
 </footer>