Cebian · Open-Source AI Side Panel Agent Edge extension security report

Cebian is an open-source AI agent that lives in your browser's side panel. Bring your own API key from OpenAI, Anthropic, Google, or any OpenAI-compatible provider — no account, no subscription, no middleman. Your chats, prompts, and files stay in your browser; only the messages you actually send leave, and they go straight to the provider you picked. It's a real agent, not a chat box: it can read the page you're on, pick any element off the screen as context, call external tools through MCP, run reusable Skills, attach screenshots and files, and switch between models mid-conversation without losing context. ──────────────────────────────────────── KEY FEATURES ──────────────────────────────────────── • Page-aware chat — Ask about the article, docs page, or search result you're looking at. The AI sees the page title, URL, and full text as context. • Element picker — Click any element on the page (a table, a code block, an error message) to attach it precisely to your next message. • Slash prompts — Build a personal library of prompt templates with variables like {{selected_text}}, {{page_url}}, {{clipboard}}. Trigger with /sum, /translate, and so on. • Skills — Package multi-step workflows or domain knowledge into reusable Skills the agent invokes on demand. Local files, importable and exportable. • MCP tools — Built-in Model Context Protocol client. Plug in a local MCP server (database, filesystem, internal API) and the agent can call it from chat. • Mobile-device emulation — Toggle responsive-design emulation on the active tab with one click. Uses DevTools Protocol, no separate setup. • Multi-model — OpenAI, Anthropic, Google, GitHub Copilot (via OAuth, reuses your existing subscription), OpenAI Codex CLI, and any OpenAI-compatible endpoint (Ollama, LM Studio, Azure OpenAI, vLLM, OpenRouter, ...). Switch providers mid-session. • Attach anything — Drag, drop, paste, or upload images and files into any message. ──────────────────────────────────────── WHO IT'S FOR ──────────────────────────────────────── • Developers — ask about a GitHub issue, a Stack Overflow answer, or a docs page in place; attach screenshots of UI bugs; toggle mobile emulation without opening DevTools. • Writers and researchers — summarize, translate, or rewrite anything you're reading without leaving the page. • People who already pay for multiple AI providers — pick the right model per task instead of being locked into one. • Privacy-conscious users — nothing leaves your machine except the messages you explicitly choose to send. ──────────────────────────────────────── PRIVACY & OPEN SOURCE ──────────────────────────────────────── Cebian has no backend. Chat history, API keys, prompt templates, Skills, attachments, and MCP configurations all live in your browser's local storage. The only outbound network traffic is the message content you send, going directly to the AI provider you configured under that provider's privacy terms. Licensed under AGPL-3.0. Full source code on GitHub. Uninstall the extension and every byte of your data is gone — by design. ──────────────────────────────────────── SUPPORTED AI PROVIDERS ──────────────────────────────────────── OpenAI · Anthropic · Google Gemini · GitHub Copilot (OAuth) · OpenAI Codex CLI (OAuth) · any OpenAI-compatible endpoint (Ollama, LM Studio, Azure OpenAI, vLLM, OpenRouter, ...) ──────────────────────────────────────── LINKS ──────────────────────────────────────── Source code: https://github.com/maotoumao/Cebian Issues & ideas: https://github.com/maotoumao/Cebian/issues