ETA — Edge Thin Agent

A pure browser-based AI Agent client — no installation, no backend, no Docker, no Python required. Just open the HTML file and start chatting.

Features

• Zero Setup — Pure frontend, runs entirely in the browser. No server, no Docker, no Python, no Node.js. Just open index.html in any modern browser.
• Multi-Model Support — Switch between models (Claude, GPT, etc.) on the fly. Fetch available models from your API provider or add custom ones.
• Agent Web Search — Built-in agent with internet access: Brave Search, arXiv, Semantic Scholar, OpenAlex, CrossRef, PubMed, Wikipedia, Hacker News, Stack Overflow, GitHub search, and webpage scraping. Up to 20 rounds of autonomous search per query. Only Brave requires a key; it is sent as a request header directly to Brave, never through a third-party proxy.
• Conversation Tree — Branch, edit, and rewind conversations. Navigate between branches with ease.
• Knowledge Buffer — Right-side panel for caching search results, uploaded files, and fetched web pages. The agent reads from it on demand.
• Long-term Memory — A small set of durable facts about you, shared across conversations. The agent decides what is worth keeping (stable preferences and habits only, never task details), with a hard character cap; going over triggers one LLM call that merges and shortens entries. Inspect, edit or delete entries individually via 🧠 in the top bar, or turn the whole thing off.
• Personal Info — A fixed profile you fill in yourself (name, role, tech stack, tone and format preferences) under 👤 in the top bar. Read-only to the model and never auto-compressed.
• File Upload — Supports images (inline), PDF, DOCX, PPTX, XLSX, and various text/code files.
• Markdown + Math — Full Markdown rendering with syntax highlighting (highlight.js) and LaTeX math (KaTeX).
• Thinking Models — Supports <think> tag parsing for chain-of-thought models, with collapsible thinking blocks.
• Context Compression — Automatically summarizes long conversations (>120k chars) via LLM to stay within context limits.
• Dark / Light Theme — Toggle between dark and light UI in Settings.
• Bilingual UI — Switch between Chinese and English interface in Settings.
• Export — Export single conversations as Markdown or all conversations as JSON.
• Balance Check — Query your API provider's remaining balance.
• Debug Panel — Real-time agent loop logs with console interception.
• Data Privacy — All data stored locally in your browser (IndexedDB, with settings in localStorage), including long-term memory and your profile. Nothing is sent anywhere except your configured API endpoint.

Quick Start

1. Clone or download this repository.
2. Open index.html in your browser.
3. First time? A setup dialog will appear — enter your Base URL and API Key.
4. Start chatting!

Note: Config is saved to browser localStorage. If you move the folder to a new location, the setup dialog will appear again for you to re-enter your keys. Optionally, fill js/env.js to auto-populate keys (this file is git-ignored).

Required API Keys

See js/env.js.example for the full list of keys and where to apply for them.

License

MIT