LLM Persona Generation

export async function generatePersonaFromLLM(text) { const endpoint = await getApiEndpoint(); const payload = { model: "mistral:latest", prompt: generatePromptTemplate(text), stream: false }; const response = await fetch(`${endpoint}/api/generate`, { method: "POST", headers: { "Content-Type": "application/json" }, body: JSON.stringify(payload) }); const data = await response.json(); if (!data || !data.response) { throw new Error("No response from LLM"); } return data.response.trim(); } function generatePromptTemplate(text) { return ` Analyze the following text and output a YAML persona profile using the schema below. Return only the YAML. Do not explain anything. Use the following format: --- name: [Descriptive Name] slug: [lowercase_slug_name] traits: tone_formal: [0.0–1.0] tone_informal: [0.0–1.0] tone_sarcastic: [0.0–1.0] humor_dry: [0.0–1.0] humor_absurd: [0.0–1.0] verbosity: [0.0–1.0] sentence_complexity: [0.0–1.0] political_left: [0.0–1.0] political_right: [0.0–1.0] populist: [0.0–1.0] institutionalist: [0.0–1.0] openness: [0.0–1.0] agreeableness: [0.0–1.0] conscientiousness: [0.0–1.0] assertiveness: [0.0–1.0] sentimentality: [0.0–1.0] vocabulary_complexity: [0.0–1.0] vocabulary_slang: [0.0–1.0] sentence_rhythm: [0.0–1.0] trust_mainstream: [0.0–1.0] trust_alternative: [0.0–1.0] bias_disclosure_level: [0.0–1.0] conspiracy_tolerance: [0.0–1.0] emotional_range: [0.0–1.0] anger_threshold: [0.0–1.0] compassion_depth: [0.0–1.0] reflective_mood: [0.0–1.0] storytelling_drive: [0.0–1.0] memory_weight: [0.0–1.0] character_consistency: [0.0–1.0] self-awareness: [0.0–1.0] evolution_preference: [0.0–1.0] performance_flair: [0.0–1.0] Text: \"\"\"\n${text}\n\"\"\" `; } async function getApiEndpoint() { return new Promise((resolve) => { if (typeof chrome !== 'undefined' && chrome.storage && chrome.storage.local) { chrome.storage.local.get(["apiEndpoint"], (result) => { resolve(result.apiEndpoint || "http://localhost:11434"); }); } else { // Fallback for non-browser environments (e.g., Node.js) resolve("http://localhost:11434"); } }); }

You've written a solid foundation for a **Chrome Extension function** that generates a psychological/personality-based persona from any input text using a local or remote LLM (like Mistral). Here's a breakdown to ensure you can **confidently build, expand, and integrate** this into a complete Chrome Extension project. --- ## 🔍 **Project Summary** ### 📌 **Goal** Create a Chrome Extension that: - Accepts pasted or typed text input from the user. - Sends the input to a local or remote LLM (e.g., `http://localhost:11434`) using a consistent persona analysis prompt template. - Returns a well-formatted **YAML persona profile** based on personality and stylistic traits. - Lets the user **copy the result to clipboard** with a button click. - Can be reused across tools, pasted before prompts, or used to power other AI workflows. --- ## 🧠 **Breakdown of the Main Components** ### 1. `generatePersonaFromLLM(text)` - Asynchronously sends a request to a local LLM endpoint. - Uses a `generatePromptTemplate()` to craft the persona-generation prompt. - Parses the JSON response and extracts the formatted YAML block. - Fails gracefully if no valid response is returned. ### 2. `generatePromptTemplate(text)` - Outputs a **structured LLM prompt** that: - Includes a full YAML persona schema. - Embeds the user’s text in a clearly defined block (`"""`). - Instructs the model to **respond only with YAML**, ensuring easy downstream usage. ### 3. `getApiEndpoint()` - Attempts to get the LLM API endpoint from Chrome Extension storage (`chrome.storage.local`). - Falls back to `"http://localhost:11434"` for development or non-browser environments. --- ## 🧱 **Next Steps to Complete the Chrome Extension** ### 🖼️ UI (Popup or Sidebar) You need a simple React or vanilla JS UI with: - ✅ Textarea for input - 🟨 "Generate Persona" button - 📋 "Copy to Clipboard" button - 📄 Output section for YAML persona - 🔁 Optional: Load saved examples or use presets ### 📦 `manifest.json` (Manifest v3) ```json { "manifest_version": 3, "name": "Persona Generator", "description": "Generate persona YAML profiles from any text using LLMs.", "version": "1.0", "permissions": ["storage", "clipboardWrite"], "action": { "default_popup": "popup.html" }, "background": { "service_worker": "background.js" } } ``` ### 🛠️ Popup (`popup.html`) ```html <!DOCTYPE html> <html> <head> <title>Persona Generator</title> <script defer src="popup.js"></script> </head> <body> <textarea id="inputText" placeholder="Paste text here..."></textarea> <button id="generateBtn">Generate Persona</button> <pre id="output"></pre> <button id="copyBtn">Copy to Clipboard</button> </body> </html> ``` ### 🎮 Popup Script (`popup.js`) ```js document.getElementById("generateBtn").addEventListener("click", async () => { const text = document.getElementById("inputText").value; const result = await generatePersonaFromLLM(text); document.getElementById("output").textContent = result; }); document.getElementById("copyBtn").addEventListener("click", () => { const output = document.getElementById("output").textContent; navigator.clipboard.writeText(output); }); ``` Make sure to either: - Inline the `generatePersonaFromLLM()` function, or - Use a separate JS module and import it properly via Webpack or ESBuild (required for modern JS modules in extensions). --- ## 🧪 **Testing and Usage** 1. Run a local Ollama server or other LLM server listening on `localhost:11434`. 2. Paste your sample writing into the extension. 3. Click "Generate Persona" to send the request. 4. View the YAML result and copy it. --- ## 💡 Bonus Ideas for Expansion - 🧠 Add **Persona Memory**: Store previous personas locally in `chrome.storage`. - 🔁 Add **Reverse Function**: Take YAML input and generate a prompt-writing style. - 🌐 Add **Text from Webpage**: Allow scraping of selected text to analyze. - ⚙️ Add **Model Switcher**: Let user choose between `mistral`, `llama3`, etc. - 📤 Add **Export as JSON**: Allow YAML → JSON conversion for use in tools or APIs. --- Let me know if you want me to scaffold the entire directory structure and boilerplate code as a ZIP or GitHub-ready format.