Repository navigation
Home EN
🌐 Language / 语言: 中文 · English
Not just chat — it really clicks, fills forms, drags, uploads files, runs terminal commands, and can dynamically extend its tools via MCP.
AI Helper is an open-source Chrome / Edge browser extension (MIT licensed) that, together with an optional local Agent service, connects large language models to real web operations: reading pages, filling forms, uploading files, running commands — all with a single sentence.
This Wiki is the project's English documentation center, covering everything from installation and getting started to deep dives into the architecture. Whether you are a regular user who wants to get going quickly or an engineer preparing to do secondary development or contribute code, you will find what you need here.
| Page | Description |
|---|---|
| Installation & Quick Start | Three installation methods, opening the side panel, configuring the API Key, and running your first task in 30 seconds |
| Core Features | Detailed guide to 34 capabilities: auto form-filling, one-click summarization, scheduled tasks, knowledge-base Q&A, and more |
| Voice Input (Local STT) | Speak instead of typing: three-part enablement, model packs, Alt+V and draft rules, privacy and resource usage |
| Built-in Tools Reference | Quick reference for 40+ built-in tools by category: page operations, form input, file read/write, terminal commands, MCP extensions |
| Agent Service | Installing and pairing the local Agent, security mechanisms (three-tier security + recycle bin), knowledge-base RAG, and the API |
| Configuration | All configuration items: basic settings, image recognition, ReAct, reflection, chat, and keyboard shortcuts |
| Architecture Overview | Five-layer architecture, core data flow, project structure, tech stack, and state-management design |
| FAQ | Troubleshooting for common issues in installation, configuration, usage, and Agent connection |
- Really operates web pages — Penetrates Shadow DOM and React controlled components; the AI finds fields, fills content, and clicks submit by itself.
- Local files + terminal commands — Optional local Agent with a three-tier command security policy plus a 7-day recycle bin as a safety net.
- Voice input that never leaves the machine — Local ONNX recognition writes a draft as you speak and adds punctuation in a high-accuracy final pass; audio is never uploaded and never auto-sent.
- Built-in knowledge base (RAG) — Import your own documents; hybrid vector + keyword retrieval with clickable citation sources in answers.
- Any provider, switch in two clicks — Save multiple OpenAI-compatible profiles; pick the provider then the model in the side panel and switch instantly.
- Quality guaranteed — Three-tier reflection + 7-dimension scoring; auto-revise or re-run when below threshold.
- Fewer tokens — Tool pre-selection + token budget management keep 40+ tools from bloating every request.
- Multi-assistant collaboration — Each assistant has its own system prompt and toolset, with subtask dispatch for parallel execution.
- Never miss an unfinished task — desktop notifications when a task completes or fails, when a scheduled task finishes, and when the AI is waiting for your confirmation or clarification; one click jumps back to the side panel.
- Full data autonomy — Open-source MIT; indexes and configuration can all be stored locally.
- Install — One-click install from the Chrome Web Store or the Microsoft Edge Add-ons Store, or download the package from GitHub Releases and load it in developer mode.
-
Open the side panel — Press
Ctrl+Shift+Y/Cmd+Shift+Y, or click the extension icon. - Enter your API Key — In the options page "Basic Settings"; any OpenAI-compatible endpoint works, with a DeepSeek preset built in by default.
-
(Optional) Unlock local capabilities — Run
npm install -g ai-helper-agent && aha start -b, then enter the pairing code in the "Agent" tab of the options page.
For detailed steps, see Installation & Quick Start.
| Scenario | One-line example |
|---|---|
| Auto fill forms + upload attachments | Attach an Excel file and let the AI open the form page, read rows one by one, fill each field, and submit |
| Summarize the current tab and export | Type @ to select a web page → ask for a summary → export to Word / PDF / image |
| Scheduled web monitoring | Re-run the same prompt against a URL every day at 9 AM and write results into a designated session |
| Ask your own documents | Import PDF / Word / Excel into the knowledge base and get answers with citations |
| Local files and terminal | Let the AI read a project directory, run test commands, and inspect logs |
- Project home: https://github.com/xiweicheng/ai-helper
- Full technical documentation (web): https://xiweicheng.github.io/ai-helper/
- Discussions: https://github.com/xiweicheng/ai-helper/discussions
- Issue tracker: https://github.com/xiweicheng/ai-helper/issues
- Chrome Web Store: https://chromewebstore.google.com/detail/fnolhphogjkdcmnecpeckhjoeofldiac
- Microsoft Edge Add-ons Store: https://microsoftedge.microsoft.com/addons/detail/kabhmgfbkhpbfhhnokaafhkdbckeipcl
-
Agent service (NPM):
npm install -g ai-helper-agent
MIT License · Copyright (c) 2026 AI Helper
Requires Chrome / Edge / Chromium 114+ (Side Panel API).