parchi

0xSeroMIT活跃维护

装在浏览器里的 AI 助手,能直接操作网页完成自动化任务,不绑定特定模型,适合把日常浏览和重复操作交给 Agent 处理。

Your AI friend right in your browser

v0.6.0发布于 2026-03-23 20:02 UTC546star58fork同步于 2026-11-10 23:56 UTC

怎么上手

由 README 提炼,执行前请确认
git clone https://github.com/0xSero/parchi.git
cd parchi
npm install
npm run build
硬件要求按 README 说明

README 未说明本地硬件要求;模型可接本地 Ollama/LM Studio(http://localhost:11434/v1)或云端 API,显存需求取决于所选模型。

下载:源码 zip·GitHub Releases(安装包)本站不镜像安装包

项目说明

由 README 自动翻译提炼

装在浏览器侧边栏的 AI 助手,用对话驱动网页导航、点击、填表和信息提取

装在浏览器里的 AI 助手,能直接操作网页完成自动化任务,不绑定特定模型,适合把日常浏览和重复操作交给 Agent 处理。

核心能力

  • 在侧边栏流式对话并展示工具执行时间线
  • 提供导航、读取、交互、标签页、截图等浏览器工具
  • 按 profile 配置模型与 provider,支持多套切换
  • orchestrator 加 subagent 的多智能体执行流程
  • 工具权限与域名白名单控制,可限制自动化范围

容易踩的坑

构建后需在 chrome://extensions 开启开发者模式并加载 dist/ 目录,不是直接安装商店包;Firefox 要单独跑 npm run build:firefox 并加载 dist-firefox/;使用前必须在扩展设置里填好 OpenAI 兼容的 API key、base URL 和模型 ID。

内容由模型根据 README 生成,可能不完整或不准确,以官方文档为准。看原文 README

最近发版

5 个版本v0.6.0
  1. v0.6.0v0.6.0 — UI Polish & Settings Cleanup最新正式版2026-03-23 · 2026-03-23 20:02 UTC

    What's new

    • Text zoom controls moved from composer toolbar to Settings > Advanced > Look & Feel (with A-/A+ buttons + slider)
    • Plan drawer minimized by default — plans start collapsed, expand on user click
    • Interleaved thinking — reasoning blocks appear inline with content flow instead of pinned to the top
    • Composer tightened — reduced bottom padding for more vertical space
    • Advanced settings reorganized — Look & Feel first (open by default), logical section ordering, Danger Zone collapsed by default
    • TypeScript fix for queued message banner null check
  2. v0.5.12026-03-14 · 2026-03-14 08:43 UTC
    查看更新说明
  3. v0.5.02026-03-10 · 2026-03-10 19:30 UTC
    查看更新说明

    Parchi v0.5.0 release

    Changes:

    • Provider model visibility and paid mode gating fixes
    • Code refactoring: split oversized files to satisfy repo-standards line limits
    • Various improvements and bug fixes
  4. v0.4.62026-02-24 · 2026-02-24 00:04 UTC
    查看更新说明

    Highlights

    • Fixed managed paid runtime auth flow for OpenRouter proxy runs (including refresh/retry handling).
    • Corrected OpenRouter proxy route usage to /ai-proxy/openrouter/v1 to prevent invalid JSON/runtime failures.
    • Improved sidepanel paid UX: compact paid badge, hover/meta status details, and clearer runtime health signaling.
    • Hardened session tab favicon handling with safer fallbacks (no broken icon behavior).
    • Added richer Account & Billing usage visualization (credit flow, 7-day spend bars, runtime density).
    • Built and attached Firefox package: parchi-0.4.6.xpi.
  5. v1.0.12026-02-17 · 2026-02-17 19:58 UTC
    查看更新说明

    Convex backend + account/payments integration, OAuth settings tab, and Kimi paid proxy routing.

用户点评

登录用户的真实使用记录,可按版本和硬件筛选
共 0 条

    还没有人点评。第一个说出真实体验的人最有价值。

    README 原文

    未翻译

    Parchi

    AI-powered browser copilot (Chrome/Firefox extension).

    Chat-driven browser automation in a side panel: navigate, read, click, type, extract, and summarize.

    Chrome MV3 Firefox License: MIT


    ⚠️ Safety Notice

    Browser automation can perform sensitive actions on your behalf. Use this tool only if you understand and accept the risks:

    • automation may violate site terms of service
    • prompt-injection can cause unsafe actions
    • sensitive information can be exposed if you run untrusted prompts/pages

    Always review model output and keep strict tool/domain controls enabled when needed.


    📦 Installation

    Chrome (recommended)

    git clone https://github.com/0xSero/parchi.git
    cd parchi
    npm install
    npm run build
    

    Then:

    1. Open chrome://extensions
    2. Enable Developer mode
    3. Click Load unpacked
    4. Select dist/

    Firefox

    npm run build:firefox
    

    Then load the generated extension from dist-firefox/ in about:debugging.


    🔧 Configure a Model

    Open the extension settings and configure an OpenAI-compatible provider:

    • OpenAI https://api.openai.com/v1
    • Anthropic-compatible
    • OpenRouter https://openrouter.ai/api/v1
    • Local (Ollama/LM Studio) http://localhost:11434/v1
    • any other OpenAI-compatible endpoint

    You can set:

    • API key
    • endpoint/base URL
    • model ID
    • optional custom headers

    ✨ Core Features

    • streaming chat + tool execution timeline
    • browser tools (navigate, read, interact, tabs, screenshots)
    • profile-based model/provider configuration
    • orchestrator + subagent flow
    • session history and exports
    • tool permissions and domain allowlist controls

    🏗 Repo Structure

    PathPurpose
    packages/extension/Browser extension runtime + UI + tools
    packages/backend/Convex backend (auth/billing/proxy)
    packages/shared/Shared schemas, prompts, runtime types
    scripts/Build/release/check scripts
    tests/Unit/integration/e2e/orchestrator/perf harnesses

    🔨 Development

    npm install
    npm run build
    npm run typecheck
    npm run lint
    npm run test:unit
    npm run test:integration
    npm run test:e2e
    npm run check:repo-standards
    

    If you change sidepanel UI code, rebuild and reload the extension from dist/.


    📄 License

    MIT