Show HN: Ember – Lightweight headless browser for AI agents (17MB idle)

  ███████╗███╗   ███╗██████╗ ███████╗██████╗ 
  ██╔════╝████╗ ████║██╔══██╗██╔════╝██╔══██╗
  █████╗  ██╔████╔██║██████╔╝█████╗  ██████╔╝
  ██╔══╝  ██║╚██╔╝██║██╔══██╗██╔══╝  ██╔══██╗
  ███████╗██║ ╚═╝ ██║██████╔╝███████╗██║  ██║
  ╚══════╝╚═╝     ╚═╝╚═════╝ ╚══════╝╚═╝  ╚═╝

Open source, lightweight headless browser for AI agents.

pip install ember-browser

No Docker. No API key to start.

Most web tools for agents ship with Chromium (~281 MB) or require Docker just to get started. We needed something an agent could use on a VPS, a laptop, or a Raspberry Pi without thinking about it.

ember runs at ~17 MB idle. It decides whether a page needs a browser — you just pass it a URL.

pip install ember-browser
ember version                  # verify install

ember                          # start the interactive session
ember url https://example.com  # or run a one-shot command
ember serve                    # start the REST API

ember

with no arguments opens a persistent session. Commands and a save guide are shown on startup — no need to type help

first.

  ███████╗███╗   ███╗██████╗ ███████╗██████╗
  ██╔════╝████╗ ████║██╔══██╗██╔════╝██╔══██╗
  █████╗  ██╔████╔██║██████╔╝█████╗  ██████╔╝
  ██╔══╝  ██║╚██╔╝██║██╔══██╗██╔══╝  ██╔══██╗
  ███████╗██║ ╚═╝ ██║██████╔╝███████╗██║  ██║
  ╚══════╝╚═╝     ╚═╝╚═════╝ ╚══════╝╚═╝  ╚═╝

  v0.1.0  lightweight headless browser for AI agents

  url        <url>              scrape a page to markdown
  search     <query>            web search
  crawl      <url>              crawl a whole website
  map        <url>              discover all URLs on a site
  interact   <url>              control a browser with natural language
  extract    <url>              pull structured data with an LLM
  batch      <urls.txt>         scrape many URLs concurrently

  ─── saving results ───────────────────────────────────────────
  one result   url example.com -o page.md
  everything   output ./research/  then all results auto-save
  last result  save page.md        after any command

ember › url andausman.com
ember › save page.md

ember › output ./research/       # auto-save everything from here
ember/research › search "python asyncio" -n 10
ember/research › crawl docs.example.com
ember/research › output clear    # stop auto-saving
ember › quit

Every command works standalone too:

ember url https://example.com                         # scrape a page
ember search "AI agents python" -n 10                 # web search
ember crawl https://docs.example.com --max-pages 20   # crawl a site
ember map https://example.com                         # discover all URLs
ember interact https://amazon.com \
  --prompt "find a mechanical keyboard under $100"
ember extract https://example.com/pricing \
  --prompt "list all plans and prices as JSON"

All commands accept -o

to save that run:

ember url https://example.com -o page.md
ember search "python" -o results.json
ember crawl https://docs.example.com -o ./pages/   # one .md per page
ember map https://example.com -o urls.txt
ember extract https://example.com -o data.json

Set a default save directory so you never need -o

:

ember config --save-dir ./research/    # persists across sessions
ember config                           # show current settings
ember config --save-dir ""             # clear it

Or use an environment variable for the current shell:

EMBER_SAVE_DIR=./out ember url https://example.com

In a session, the three ways to save:

ember › url example.com -o page.md     # save just this run
ember › save page.md                   # save the last result
ember › output ./research/             # auto-save all results from now on
ember batch urls.txt                      # 5 concurrent by default
ember batch urls.txt -c 20 -o ./pages/   # 20 parallel, save to dir
python
from emb.scrape import scrape_url, scrape_markdown
from emb.search import search
from emb.crawl import crawl
from emb.map import map_url

result = scrape_url("https://example.com")
print(result.markdown)   # full page content as markdown
print(result.title)      # page title
print(result.success)    # True / False

md = scrape_markdown("https://example.com")

result = crawl("https://docs.example.com", max_pages=20, max_depth=3)
for page in result.pages:
    print(page.url, len(page.markdown))

result = map_url("https://example.com", max_links=100)
print(result.links)   # list[str]

results = search("python asyncio tutorial", limit=5)
for r in results:
    print(r.title, r.url)

from emb.interact import interact

result = interact("https://example.com", prompt="click the login button")
print(result.content)   # what the agent did / saw

from emb.agent import extract

data = extract("https://example.com/pricing", prompt="list all plans and prices")
print(data)   # dict
python
import asyncio
from emb.scrape import scrape_url_async

async def main():
    results = await asyncio.gather(
        scrape_url_async("https://example.com"),
        scrape_url_async("https://httpbin.org/get"),
    )
    for r in results:
        print(r.url, r.success)

asyncio.run(main())
ember serve               # http://127.0.0.1:51251
ember serve --port 8080   # custom port

EMBER_API_KEY=your-secret ember serve   # require auth
curl -X POST http://localhost:51251/scrape \
  -H "Content-Type: application/json" \
  -H "X-API-Key: your-secret" \
  -d '{"url": "https://example.com"}'

curl -X POST http://localhost:51251/search \
  -H "Content-Type: application/json" \
  -d '{"query": "AI agents", "limit": 5}'

curl -X POST http://localhost:51251/crawl \
  -H "Content-Type: application/json" \
  -d '{"url": "https://docs.example.com", "max_pages": 10}'

Endpoints: /scrape

/search

/crawl

/map

/interact

/extract

/agent

/health

Add to your Hermes config, OpenClaw config, Mercury config, or any MCP-compatible host:

{
  "mcpServers": {
    "ember": {
      "command": "ember",
      "args": ["mcp"]
    }
  }
}

Works with Hermes, OpenClaw, Mercury, and any MCP-compatible host.

Available tools: scrape

, search_web

, crawl_site

, map_site

, batch_scrape

, interact_page

, extract_data

.

Once connected, your agent can use ember tools directly in conversation:

User: Summarise the latest posts on Hacker News

Agent: [calls scrape("https://news.ycombinator.com")]
       → returns full page markdown with titles, scores, links

Agent: Here are today's top stories on Hacker News: ...
User: Find 5 articles about AI agents and scrape each one

Agent: [calls search_web("AI agents 2025", limit=5)]
       → returns list of {title, url, description}

Agent: [calls batch_scrape(["url1", "url2", ...])]
       → returns markdown for each page

Agent: Here's a summary across all 5 articles: ...

Not every page needs a browser. ember knows the difference.

Tier 1 — trafilatura handles ~89% of the web: blogs, news, documentation, docs sites, GitHub. Pure HTTP, no browser process, no memory overhead.

Tier 2 — Lightpanda handles JavaScript-heavy pages, SPAs, and interactive content. It's a real browser engine written in Zig, built for machines rather than humans — 20 MB total. ember downloads and caches it automatically on first use, and only falls back to it when tier 1 produces thin content.

Most requests never reach the browser.

Firecrawl needs 4–8 GB in Docker. Crawl4AI imports at 171 MB before scraping anything. ember fits where your agent already runs.

MIT — open source forever.