aread

AI-first web reader & search with AI summarization. The a stands for AI — built for agents that need structured web content in one call.

aread https://example.com              # read any page as Markdown
aread example.com --summarize          # read + AI summary
aread -s "rust async tutorial"         # search the web
aread -s "react hooks" --read          # search + read all results
aread -s "LLM evaluation" --multi      # query all engines, merge & deduplicate
aread config init                      # setup AI summarization

• Zero runtime dependencies (optional turndown for local fallback)
• AI summarization via OpenAI-compatible APIs (OpenAI, Groq, OpenRouter, xAI, SiliconFlow, LongCat)
• Zhihu article support with auto browser cookie extraction (macOS)
• Works on Linux, macOS, Windows
• Single file, structured output designed for LLM consumption

Why aread?

AI agents need web content as clean, structured text — not HTML soup. aread gives agents:

• One-call web reading — any URL to Markdown, handling JS rendering and anti-scraping
• AI summarization — distill pages to key facts via any OpenAI-compatible API
• Multi-source search — query multiple engines in parallel, get deduplicated results
• Zhihu support — auto-extract browser cookies to read zhihu.com articles and answers
• Structured output — --json for machine parsing, --raw for pipe-friendly text
• Graceful degradation — engine failures are silently skipped, partial results still returned

Install

npm i -g aread-cli       # or: bun i -g aread-cli

Or run directly:

npx aread-cli https://example.com

Read

Turn any web page into clean Markdown. Powered by Jina Reader with automatic local fallback.

aread https://example.com           # full page as Markdown
aread example.com                   # auto-adds https://
aread -o page.md example.com        # save to file
aread -r example.com | head -50     # raw mode, no status messages
aread --json example.com            # structured JSON output

If Jina is unreachable, aread falls back to local fetch + turndown (install with npm i turndown for fallback support).

Results are cached locally (~/.cache/aread/) with 24h expiry. Use --no-cache to bypass.

Jina Headers

Fine-tune what you get back:

aread -H "X-Target-Selector:article" https://blog.example.com    # extract <article> only
aread -H "X-With-Links:true" https://example.com                 # include hyperlinks
aread -H "X-No-Cache:true" https://example.com                   # bypass Jina cache

| Header | Description | |--------|-------------| | X-Return-Format | html, text, screenshot, pageshot | | X-With-Links | include hyperlinks | | X-With-Images | include images | | X-No-Cache | bypass Jina cache | | X-Target-Selector | extract specific CSS selector |

AI Summarization

Summarize any fetched content using an OpenAI-compatible AI API. Preserves 90% of information — only strips filler, repetition, and noise.

Setup

Interactive setup (recommended):

aread config init

Or manual configuration:

aread config set ai.provider openrouter      # choose provider
aread config set ai.apiKey sk-xxx            # set API key
aread config set ai.model google/gemini-2.0-flash-001  # set model
aread config set ai.autoSummarize true       # auto-summarize every fetch

Built-in Providers

| Provider | Base URL | |----------|----------| | openai | https://api.openai.com/v1 | | groq | https://api.groq.com/openai/v1 | | openrouter | https://openrouter.ai/api/v1 | | xai | https://api.x.ai/v1 | | siliconflow | https://api.siliconflow.cn/v1 | | longcat | https://api.longcat.chat/openai/v1 |

Or use any OpenAI-compatible endpoint:

aread config set ai.baseUrl https://your-api.com/v1

Usage

aread example.com -S               # one-time summarize with -S flag
aread example.com --summarize      # same as -S
aread example.com --no-summarize   # disable auto-summarize for this call

With ai.autoSummarize set to true, every fetch automatically includes an AI summary.

Config Management

aread config show                  # view current config (secrets masked)
aread config providers             # list all built-in providers
aread config get ai.model          # get a specific value
aread config delete ai.apiKey      # delete a value

Zhihu Support

aread can fetch articles and answers from zhihu.com (知乎) — a site with aggressive anti-crawling protection.

On macOS, aread automatically extracts cookies from your Chrome/Arc/Edge/Brave browser. Just make sure you're logged in to zhihu.com in your browser.

aread https://zhuanlan.zhihu.com/p/696916846     # article — auto works
aread https://www.zhihu.com/question/123456       # question + all answers

Supported page types:

| Type | URL Pattern | What's extracted | |------|-------------|------------------| | Article | zhuanlan.zhihu.com/p/xxx | Title, author, date, full content | | Question | zhihu.com/question/xxx | Question + all answers (sorted by votes) | | Answer | zhihu.com/question/xxx/answer/yyy | Specific answer with question title |

Manual cookie configuration (if auto-extraction doesn't work):

aread config set zhihu.cookie "<cookie from browser DevTools>"

Search

Search the web with multiple engine support. No API keys, no tracking.

aread -s "rust async tutorial"                # search (default: auto engine)
aread -s "node.js streams" -n 5              # top 5 results
aread -s "react hooks" -e bing               # use specific engine
aread -s "kubernetes" --json                  # JSON output for agents

Note: Use quotes for multi-word queries: aread -s "query with spaces".

Engines

| Engine | Flag | Notes | |--------|------|-------| | DuckDuckGo | -e duckduckgo | Private, no tracking | | Bing | -e bing | Reliable, good international coverage | | Auto | -e auto (default) | Probes DuckDuckGo, falls back to Bing |

If the chosen engine fails, aread automatically tries fallback engines before giving up.

Multi-Engine Search (--multi)

Query all engines concurrently and get merged, deduplicated results.

aread -s "WebAssembly" --multi               # query all engines at once
aread -s "transformer architecture" -m       # -m shorthand
aread -s "bilibili" --multi --json           # JSON output

Search + Read

Search the web, then read every result page as Markdown — all in one command.

aread -s "CSS grid tutorial" --read                    # search + read all
aread -s "kubernetes networking" --read -o research.md  # save everything
aread -s "SQLite vs PostgreSQL" --read -n 3             # top 3, read all
aread -s "LLM agents" --multi --read                    # multi-engine + read

With AI summarization enabled, search + read outputs only the AI summary for each result — saving tokens while preserving key information.

Agent Integration

aread outputs clean, structured content designed for LLM consumption:

# Feed a page to an AI agent
aread -r https://docs.python.org/3/tutorial | claude "summarize this"

# Multi-source research in one command
aread -r -s "WebAssembly 2024" --multi --read | claude "explain the key trends"

# Structured JSON for programmatic use
aread -s "react hooks" --multi --json | jq '.[].url'

JSON Output

Use --json for machine-readable output:

# Search results as JSON array
aread -s "example" --json

# Read a page as JSON (includes summary field when AI is configured)
aread --json https://example.com
# { "url": "...", "markdown": "...", "summary": "...", "cached": false, "error": null }

All Options

aread <URL>                  Read a page as Markdown
aread <URL> --summarize      Read + AI summary
aread -s <QUERY>             Search the web
aread -s <QUERY> --read      Search + read top results
aread config                 Manage configuration

| Flag | Description | |------|-------------| | -o, --output <FILE> | Save to file | | -r, --raw | No status messages, pipe-friendly | | -t, --timeout <SEC> | Timeout in seconds (default: 30) | | -H, --header <K:V> | Extra Jina header (repeatable) | | -s, --search <QUERY> | Search the web (use quotes for multi-word) | | -n, --num <N> | Number of results (default: 10) | | -e, --engine <ENGINE> | Search engine: duckduckgo\|bing\|auto (default: auto) | | -m, --multi | Query all engines concurrently | | --read | Fetch each search result as Markdown | | -c, --concurrency <N> | Concurrent reads with --read (default: 5) | | -S, --summarize | AI summarize (requires config) | | --no-summarize | Disable auto-summarize | | --no-cache | Skip URL cache | | --json | Structured JSON output |

How It Works

| Feature | Implementation | Cost | |---------|---------------|------| | Read | Jina Reader with local turndown fallback | Free | | Search | Native HTTP scraping of DuckDuckGo, Bing | Free | | Summarize | OpenAI-compatible API (user-provided key) | Per-token | | Zhihu | Browser cookie extraction + SSR data parsing | Free | | Cache | SHA-256 URL hashing to ~/.cache/aread/, 24h expiry | Local |

No Python. No external binaries. No API keys required (except optional AI summarization). Just Node.js 18+.

License

MIT