diff --git a/README.md b/README.md index 14adcca..ce7cb56 100644 --- a/README.md +++ b/README.md @@ -159,11 +159,10 @@ Agent Reach 做的事情很简单：**帮你把这些选型和配置的活儿做 ### 🔌 每个渠道都是可插拔的 -每个平台对应一个独立的 Python 文件（~50–100 行），实现统一接口。**后端工具随时可以换**——哪天出了更好的工具，改一个文件就行，其他不用动。 +每个平台对应一个独立的 Python 文件，实现统一接口。**后端工具随时可以换**——哪天出了更好的工具，改一个文件就行，其他不用动。 ``` channels/ -├── base.py → 统一接口（Channel 基类） ├── web.py → Jina Reader ← 可以换成 Firecrawl、Crawl4AI…… ├── twitter.py → birdx ← 可以换成 Nitter、官方 API…… ├── youtube.py → yt-dlp ← 可以换成 YouTube API、Whisper…… @@ -173,54 +172,9 @@ channels/ ├── xiaohongshu.py → mcporter MCP ← 可以换成其他 XHS 工具…… ├── rss.py → feedparser ← 可以换成 atoma…… ├── exa_search.py → mcporter MCP ← 可以换成 Tavily、SerpAPI…… -└── __init__.py → 渠道注册（加一行就注册一个新渠道） +└── __init__.py → 渠道注册 ``` -想换后端？打开对应文件，改掉 `read()` / `search()` 的实现就行。接口不变，其他代码不用动。 - -### 🧩 添加新渠道（3 步） - -添加新渠道非常简单，3 步就能搞定： - -**第 1 步：** 新建 `channels/你的渠道.py` - -```python -from .base import Channel, ReadResult, SearchResult - -class 你的渠道Channel(Channel): - name = "你的渠道" - description = "一句话描述" - backends = ["用了什么工具"] - - def can_handle(self, url: str) -> bool: - return "你的域名" in url - - async def read(self, url: str, config=None) -> ReadResult: - # 读取内容，返回 ReadResult - return ReadResult(title="...", content="...", url=url, platform=self.name) - - def check(self, config=None): - return "ok", "一切正常" - - # 可选：实现 search() 支持搜索 -``` - -**第 2 步：** 在 `channels/__init__.py` 注册 - -```python -from .你的渠道 import 你的渠道Channel - -ALL_CHANNELS: List[Channel] = [ - ... - 你的渠道Channel(), # 加这一行 - WebChannel(), -] -``` - -**第 3 步：** 没了。`agent-reach doctor` 自动识别，`agent-reach read` 自动路由。 - -> 💡 **参考现有渠道：** `rss.py`（30 行，最简单）→ `web.py`（50 行）→ `youtube.py`（100 行，含搜索） - ### 当前选型 | 场景 | 选型 | 为什么选它 | @@ -241,25 +195,9 @@ ALL_CHANNELS: List[Channel] = [ 欢迎提 [Issue](https://github.com/Panniantong/agent-reach/issues) 和 [PR](https://github.com/Panniantong/agent-reach/pulls)。 -### 🆕 想添加新渠道？ +**想要新渠道？** 直接提 Issue 告诉我们，或者自己提 PR。 -1. 复制 `agent_reach/channels/rss.py`（最简单的参考） -2. 实现 `can_handle()` + `read()`，可选 `search()` 和 `check()` -3. 在 `__init__.py` 注册 - -就这么简单。不需要改框架代码，不需要了解其他渠道。 - -**希望支持的渠道（欢迎 PR）：** - -- 📰 Hacker News — 科技新闻 -- 🐘 Mastodon / Fediverse — 去中心化社交 -- 📱 Telegram — 频道和群组 -- 🎵 Spotify / Apple Podcasts — 播客字幕 -- 📝 Medium / Substack — 付费墙文章 -- 🔬 arXiv / Semantic Scholar — 学术论文 -- 💬 Discord — 服务器消息 -- 📌 Pinterest — 图片搜索 -- …… 任何你觉得有用的平台！ +**想在本地加？** 让你的 Agent clone 下来改就行，每个渠道就是一个独立文件，加起来很简单。 ## 致谢 diff --git a/docs/README_en.md b/docs/README_en.md index eaf6247..d29482b 100644 --- a/docs/README_en.md +++ b/docs/README_en.md @@ -159,11 +159,10 @@ Agent Reach does one simple thing: **it makes those tool selection and configura ### 🔌 Every Channel is Pluggable -Each platform is a single Python file (~50–100 lines) implementing a unified interface. **Backends can be swapped anytime** — when a better tool comes along, change one file and nothing else breaks. +Each platform is a single Python file implementing a unified interface. **Backends can be swapped anytime** — when a better tool comes along, change one file and nothing else breaks. ``` channels/ -├── base.py → Unified interface (Channel base class) ├── web.py → Jina Reader ← swap to Firecrawl, Crawl4AI… ├── twitter.py → birdx ← swap to Nitter, official API… ├── youtube.py → yt-dlp ← swap to YouTube API, Whisper… @@ -173,52 +172,9 @@ channels/ ├── xiaohongshu.py → mcporter MCP ← swap to other XHS tools… ├── rss.py → feedparser ← swap to atoma… ├── exa_search.py → mcporter MCP ← swap to Tavily, SerpAPI… -└── __init__.py → Channel registry (one line to register a new channel) +└── __init__.py → Channel registry ``` -Want to swap a backend? Open the file, change the `read()` / `search()` implementation. Interface stays the same, nothing else needs to change. - -### 🧩 Adding a New Channel (3 Steps) - -**Step 1:** Create `channels/your_channel.py` - -```python -from .base import Channel, ReadResult, SearchResult - -class YourChannel(Channel): - name = "your_channel" - description = "One-line description" - backends = ["tool-name"] - - def can_handle(self, url: str) -> bool: - return "yourdomain.com" in url - - async def read(self, url: str, config=None) -> ReadResult: - # Fetch content, return ReadResult - return ReadResult(title="...", content="...", url=url, platform=self.name) - - def check(self, config=None): - return "ok", "All good" - - # Optional: implement search() for search support -``` - -**Step 2:** Register in `channels/__init__.py` - -```python -from .your_channel import YourChannel - -ALL_CHANNELS: List[Channel] = [ - ... - YourChannel(), # add this line - WebChannel(), -] -``` - -**Step 3:** Done. `agent-reach doctor` auto-detects it, `agent-reach read` auto-routes to it. - -> 💡 **Reference examples:** `rss.py` (30 lines, simplest) → `web.py` (50 lines) → `youtube.py` (100 lines, with search) - ### Current Tool Choices | Scenario | Tool | Why | @@ -239,25 +195,9 @@ ALL_CHANNELS: List[Channel] = [ [Issues](https://github.com/Panniantong/agent-reach/issues) and [PRs](https://github.com/Panniantong/agent-reach/pulls) welcome. -### 🆕 Want to Add a New Channel? +**Want a new channel?** Open an Issue to request it, or submit a PR yourself. -1. Copy `agent_reach/channels/rss.py` (simplest reference) -2. Implement `can_handle()` + `read()`, optionally `search()` and `check()` -3. Register in `__init__.py` - -That's it. No framework code to modify, no need to understand other channels. - -**Channels we'd love to see (PRs welcome):** - -- 📰 Hacker News — tech news -- 🐘 Mastodon / Fediverse — decentralized social -- 📱 Telegram — channels and groups -- 🎵 Spotify / Apple Podcasts — podcast transcripts -- 📝 Medium / Substack — paywalled articles -- 🔬 arXiv / Semantic Scholar — academic papers -- 💬 Discord — server messages -- 📌 Pinterest — image search -- … anything you find useful! +**Want to add one locally?** Just have your Agent clone the repo and modify it — each channel is a single standalone file, easy to add. ## Credits