Echo — AI Communication Copilot

Persona Case Study • Chrome Extension (Manifest V3) • Local LLMs (Ollama) & Privacy-First Vector Memory

Echo is an intelligent browser extension (an AI communication copilot) that integrates directly into your web messaging apps. It learns your personal writing style (favorite greetings, sentence length, vocabulary, and tone habits) and suggests instant, authentic replies right above your chat box — eliminating the stiff, robotic tone typical of standard AI chatbots.

1. The Problem: Why Standard AI Fails in Messaging

Most people communicating across personal, professional, and team channels face four critical frictions:

  1. The "Robotic AI" Syndrome: Standard cloud chatbots generate verbose, overly formal paragraphs ("I hope this message finds you well...") that sound completely unnatural in instant messaging.

  2. Context Switching & Latency Friction: Copy-pasting conversation threads between messaging apps and ChatGPT browser tabs breaks chat flow and takes more time than typing manually.

  3. Privacy & Corporate Data Leakage: Sending confidential work chats or personal conversations to third-party cloud servers introduces significant privacy and compliance risks.

  4. Layout Obstruction & Editor Incompatibilities: Standard browser extensions frequently overlap chat buttons or cause duplicate text insertions inside modern React, Lexical, and Slate messaging editors.

2. Supported Messaging Platforms

Echo runs natively inside Google Chrome and Chromium browsers, injecting a floating, non-intrusive frosted glassmorphic dock directly into:

  • WhatsApp Web (web.whatsapp.com)
  • Instagram Direct Messages (instagram.com/direct)
  • Telegram Web (web.telegram.org)
  • Discord Web (discord.com/app)
  • X (Twitter) DMs (x.com/messages)

3. Key Features & Capabilities

1-Click Smart Replies (<500ms Latency)

When messages arrive, Echo parses multi-turn dialogue context and generates 3 natural reply choices. Clicking an option or pressing Alt + 1 / 2 / 3 instantly inserts it into the chat input box with zero duplicate text repetitions.

Foldable & Freely Draggable Floating Toolbar

  • Left-Aligned Fold & Minimize (Alt + M): Shrinks the entire toolbar into an ultra-compact floating pill on the left so it never blocks chat history or media controls.
  • Freely Draggable Anywhere: Click and drag the ⠿ Echo handle to reposition the toolbar anywhere on your screen. Custom coordinates are persisted across page reloads.
  • Snap to Input (Dock): One-click button to snap the toolbar right back above the chat input.

Local AI-First Intelligence (Ollama llama3.2:1b & qwen3.5:4b)

  • Primary Model (llama3.2:1b): Generates contextual suggestions in under 500ms on standard laptop CPU/GPU hardware.
  • Reasoning Model (qwen3.5:4b): Handles complex threads, long-form drafting, and executive summarization.
  • Local Embeddings (nomic-embed-text): Embeds sent messages into SQLite vector storage to recall your real past replies.

1-Click Draft Polisher & Grammar Refiner (Alt + P)

Quickly draft messy thoughts with typos and refine them into 5 distinct communication styles:

  • Fix Grammar & Vocab: Corrects spelling, grammar, and phrasing while preserving your natural voice.
  • Professional: Converts casual text into crisp, articulate business English.
  • Short & Punchy: Tightens long paragraphs into direct, concise messages.
  • Casual: Relaxes tone for peer groups and friends.
  • Elaborate: Expands short bullet points into complete, well-formed thoughts.

"Catch Me Up" — Thread Summarizer (Alt + S)

Summarizes up to 30 recent messages in active conversations into 3 clear, actionable sections:

  1. Discussion Topic: Core subject and context of the ongoing conversation.
  2. Questions & Actions for You: Action items, deadlines, and pending questions requiring your response.
  3. Decisions & Next Steps: Agreed-upon conclusions and scheduling.

Voice-to-Text with AI Polish

Real-time voice dictation with automatic filler word removal ("um", "like", "you know") and grammar polishing before insertion.

Dynamic Glassmorphism Transparency

Interactive in-bar slider allowing users to adjust glass opacity from 10% (crystal clear) to 100% (solid frosted) to blend seamlessly with dark and light web themes.

Contact-Specific Relationship Memory

Learns individual relationship preferences (such as casual tone for friends, formal for managers) and automatically switches tones when switching chat threads.

3-Tier Resilient Architecture

  1. Tier 1 (Local Ollama + SQLite Vector Memory): Sub-second local inference with 100% data privacy.
  2. Tier 2 (Direct NVIDIA NIM Cloud API): High-speed cloud fallback if local daemon is offline.
  3. Tier 3 ("Own Mind" NLP Engine): Rule-based template & memory reasoning that functions completely offline without internet or an LLM.

4. Technical Architecture & Stack

Frontend / Extension UI

Chrome Extensions (Manifest V3), TypeScript, React 18, Vite, Tailwind CSS, Custom Frosted Glassmorphic CSS System, MutationObserver DOM lifecycle.

Local AI & Backend

Python 3.11, FastAPI, Uvicorn, SQLite Vector Store (Cosine Similarity Search), Local Ollama Client (llama3.2:1b, qwen3.5:4b, nomic-embed-text).

Cloud & Offline Resilience

NVIDIA NIM API (Llama 3.1 / 3.3), In-Memory Cache (45s TTL), Chrome Storage Local Persona Sync, "Own Mind" Rule-Based NLP Engine.

Editor Integration

Multi-platform DOM adapter supporting React, Lexical, Slate, Draft.js, contenteditable, and textarea inputs with zero duplication.

5. Summary in 30 Seconds

"Echo is an AI communication copilot for messaging platforms like WhatsApp and Instagram. Instead of producing generic, robotic paragraphs, it runs ultra-fast local LLMs (Llama 3.2 via Ollama) and learns your authentic writing style to provide 3 instant reply options directly above your chat box. It includes a draggable and foldable floating toolbar, fixes typos in 1 click, summarizes long threads, and keeps all personal messages 100% private on your machine."

← Back to Projects

View on GitHub →