1. Install command: git+... requires git (many envs don't have it) → Changed to zip URL: pip install .../archive/main.zip 2. yt-dlp not in dependencies (shows warning in doctor) → Added yt-dlp to required dependencies 3. search-twitter/reddit shows raw 401 error without Exa key → Now shows friendly 'Exa API key not configured' message
3.9 KiB
Agent Eyes — Installation Guide
For Humans
Copy this to your AI Agent:
Install Agent Eyes: https://raw.githubusercontent.com/Panniantong/agent-eyes/main/docs/install.md
For AI Agents
Install Agent Eyes with minimal user effort. Goal: fewest questions, fewest steps.
Step 1: Install and detect
pip install https://github.com/Panniantong/agent-eyes/archive/main.zip || pip3 install https://github.com/Panniantong/agent-eyes/archive/main.zip
agent-eyes install --env=auto
Read the output. It tells you the detected environment (local/server) and what's already working.
Step 2: Set up search (highest value, free)
"Want to unlock web search? One free API key lets me search the web, Reddit, and Twitter. Go to https://exa.ai → sign up (free, 1000 searches/month) → send me the key. Or say 'skip' — I can still read any URL you give me."
agent-eyes configure exa-key THE_KEY
Step 3: Set up cookies (the approach depends on environment)
Path A: Local computer (auto-detected)
Offer one-click import:
"I can grab your browser cookies automatically to unlock Twitter, XiaoHongShu, and Bilibili — all at once. Just close Chrome first, then say 'go'."
If user agrees:
pip install browser-cookie3 2>/dev/null; agent-eyes configure --from-browser chrome
If user uses another browser: replace chrome with firefox, edge, brave, or opera.
If it fails or user declines, fall through to Path B.
Path B: Server, or auto-import didn't work
Tell the user the easiest manual method:
"I need cookies from your browser to unlock some platforms. Here's the fastest way (30 seconds per site):
- Install the Cookie-Editor Chrome extension: https://chromewebstore.google.com/detail/cookie-editor/hlkenndednhfkekhgcdicdfddnkalmdm
- Go to the site (e.g. x.com) — make sure you're logged in
- Click the Cookie-Editor icon → Export → Header String
- Paste the result to me
Which platforms do you use? I'll tell you which ones to export."
For each platform the user wants:
Twitter/X — export from https://x.com :
agent-eyes configure twitter-cookies "PASTED_COOKIE_STRING"
(Accepts both: full cookie string OR just auth_token ct0 as two values)
XiaoHongShu — export from https://www.xiaohongshu.com :
agent-eyes configure xhs-cookie "PASTED_COOKIE_STRING"
Bilibili — export from https://www.bilibili.com : (On local computers Bilibili already works without cookies. On servers it needs a proxy — see Step 4.)
Step 4: Server-only — proxy for Reddit/Bilibili
Only mention if environment was detected as server:
"On servers, Reddit and Bilibili block direct access. Reddit search still works via Exa (free), but for reading full Reddit posts or using Bilibili, you'd need a residential proxy (~$1/month). Get one at https://webshare.io if you want. Or skip — not required."
agent-eyes configure proxy http://user:pass@ip:port
Step 5: Done
agent-eyes doctor
"✅ All set! [list active channels]. Try: 'search Reddit for ...' or 'read this URL'"
Users can configure more channels anytime — just say "set up Twitter" or "add proxy".
Quick Reference
| Command | What it does |
|---|---|
agent-eyes configure --from-browser chrome |
Auto-import ALL cookies at once (local only) |
agent-eyes configure exa-key KEY |
Unlock search (web + Reddit + Twitter) |
agent-eyes configure twitter-cookies "COOKIE_STRING" |
Unlock Twitter (cookie string or auth_token + ct0) |
agent-eyes configure xhs-cookie "COOKIE_STRING" |
Unlock XiaoHongShu |
agent-eyes configure youtube-cookies chrome |
Unlock age-restricted YouTube |
agent-eyes configure proxy URL |
Unlock Reddit reader + Bilibili (server) |
agent-eyes doctor |
Show all channels and their status |
agent-eyes read URL |
Read any URL |
agent-eyes search "query" |
Search the web |