Chromex is a Chrome MV3 side-panel assistant designed to integrate Chrome's browsing capabilities with the Codex AI platform via a local native bridge, prioritizing local credential handling outside of extension storage.

The core functionality of Chromex encompasses conversational AI interaction with diverse browser and file content. Users can chat with the current webpage, selected open tabs, screenshots, uploaded files (including PDFs, Office documents like DOCX, XLSX, XLSM, and text-based files like CSV, TSV), images, and browser history. It provides summarization capabilities for various content types, including web pages, YouTube videos (with current timestamp context and seek actions), news articles, research papers, PDFs, and arXiv documents. The system also supports image editing and generation through Codex image workflows, with local output handling. Input modalities include voice transcription, live voice mode, and attachment of supported file types. Browser control workflows are executed via Chrome content scripts, featuring visible in-page activity indicators. The user interface offers a persistent MV3 side panel with a chat-first user experience, automatic request routing, @ and / pickers for tab and profile selection respectively, page-aware suggestions, and custom profiles. Non-destructive image editing is available for uploaded images, page images, or visible screen captures. Markdown rendering with code blocks, tables, links, and copy controls is supported. Optional Codex skills can be loaded locally from a user's .codex/skills/*/SKILL.md directory when enabled.

The architectural methodology of Chromex is defined by a multi-layered runtime boundary to ensure secure and efficient communication:

1. Chrome Extension (packages/extension): This is the user-facing MV3 side-panel interface, initiating requests and displaying responses.
2. Native Messaging Host (packages/native-host): This component acts as a relay, facilitating communication between the Chrome Extension and the local bridge. Chrome's Native Messaging API is utilized for this cross-process communication.
3. Local Bridge (packages/bridge): This serves as a daemon responsible for mediating between the Native Messaging Host and the Codex app-server. It also manages multimodal workflows, including handling generated image originals, temporary uploads, and diagnostics.
4. Codex app-server: This is the underlying AI engine and API handler, processing requests and generating responses.

Security and privacy are emphasized through several default behaviors. The extension does not store sensitive credentials such as raw OpenAI API keys, OAuth tokens, or ChatGPT session tokens in Chrome extension storage. Codex OAuth and ChatGPT login processes are managed through the local Codex app-server flow. API-key login is an optional local fallback, requiring explicit user confirmation for activation. Page content, tab data, screenshots, browser history, microphone input, and browser actions are processed exclusively for user-initiated workflows. Permissions (e.g., history, tabs, screen capture, microphone, and site access) are requested only when a specific feature necessitates them. Conversation history is session-only by default, with persistent local chat history being an opt-in feature. Native-host child processes and workspace hooks operate within a reduced environment allowlist to minimize potential security risks. The extension ships with Chrome _locales entries for a wide range of languages, and model responses are instructed to adhere to the selected UI language unless otherwise specified by the user.