feat: local Kimi collection via Playwright + headless Chrome on 246

- usage_collector_kimi_local.py: Playwright connect_over_cdp to local Chrome
- kimi_login.py: interactive login helper for headless Chrome
- Replaces SSH-to-Windows approach, removes single point of failure
- Chrome CDP service: chrome-kimi-cdp.service (systemd)
This commit is contained in:
hmo
2026-07-21 00:56:00 +08:00
parent 2598b286eb
commit c11222e55c
2 changed files with 342 additions and 0 deletions
+89
View File
@@ -0,0 +1,89 @@
#!/usr/bin/env python3
"""
kimi_login.py — 在 246 headless Chrome 上登录 Kimi
====================================================
用法(需要 X11 forwarding 或手动操作):
ssh -X hmo@192.168.1.246
python3 kimi_login.py
或者无头模式下手动注入 cookies:
python3 kimi_login.py --inject-cookies '{"name":"...","value":"...","domain":".kimi.com"}'
登录成功后,cookies 自动保存在 /home/hmo/chrome-kimi-profile/ 下,
后续 chrome-kimi-cdp.service 重启后仍然保持登录态。
"""
import json
import sys
import time
from playwright.sync_api import sync_playwright
CDP_URL = "http://127.0.0.1:9222"
KIMI_URL = "https://www.kimi.com/"
def check_logged_in(page):
"""Check if Kimi is logged in by looking for user avatar or quota page access."""
try:
page.goto("https://www.kimi.com/membership/subscription?tab=quota",
wait_until="networkidle", timeout=15000)
text = page.inner_text("body") or ""
if "login" in page.url or "passport" in page.url:
return False
# If we can see usage data, we're logged in
if "%" in text:
return True
return False
except Exception:
return False
def main():
print("[kimi-login] Connecting to CDP...")
with sync_playwright() as p:
try:
browser = p.chromium.connect_over_cdp(CDP_URL)
except Exception as e:
print(f"[kimi-login] ERROR: Cannot connect to CDP: {e}")
print("[kimi-login] Is chrome-kimi-cdp.service running?")
sys.exit(1)
context = browser.contexts[0] if browser.contexts else browser.new_context()
page = context.new_page()
# Check if already logged in
print("[kimi-login] Checking current login status...")
if check_logged_in(page):
print("[kimi-login] ✅ Already logged in!")
print("[kimi-login] Cookies are saved in /home/hmo/chrome-kimi-profile/")
return
print("[kimi-login] Not logged in. Opening Kimi login page...")
print("[kimi-login] " + "=" * 50)
print("[kimi-login] If running with X11 forwarding (ssh -X), a browser window should appear.")
print("[kimi-login] Please login manually in the browser window.")
print("[kimi-login] Waiting up to 120 seconds for login...")
print("[kimi-login] " + "=" * 50)
page.goto(KIMI_URL, wait_until="networkidle", timeout=30000)
# Wait for user to login (poll every 3 seconds, up to 120s)
for i in range(40):
time.sleep(3)
current_url = page.url
print(f"[kimi-login] Checking... ({(i + 1) * 3}s) URL: {current_url[:80]}")
if "login" not in current_url and "passport" not in current_url:
# Might be logged in now
if check_logged_in(page):
print("[kimi-login] ✅ Login successful!")
print("[kimi-login] Cookies saved to /home/hmo/chrome-kimi-profile/")
return
print("[kimi-login] ❌ Login timeout (120s). Please try again.")
print("[kimi-login] Make sure X11 forwarding is enabled: ssh -X hmo@192.168.1.246")
if __name__ == "__main__":
main()
@@ -0,0 +1,253 @@
#!/usr/bin/env python3
"""
usage_collector_kimi_local.py — Kimi 用量采集(Playwright + 本地 CDP
=====================================================================
连接本地 Chrome CDP (127.0.0.1:9222, chrome-kimi-cdp.service)
打开 Kimi 控制台,提取用量数据,保存到 usage_stats_kimi.json。
替代旧的 SSH-to-Windows 方案,直接在 246 本地运行。
调用方式: python3 usage_collector_kimi_local.py
"""
import json
import os
import re
import sys
import time
from pathlib import Path
from datetime import datetime, timezone
from playwright.sync_api import sync_playwright
CDP_URL = "http://127.0.0.1:9222"
KIMI_CONSOLE = "https://www.kimi.com/membership/subscription?tab=quota"
SCRIPT_DIR = Path(__file__).resolve().parent
GATEWAY_DIR = SCRIPT_DIR.parent
TEMP_DIR = GATEWAY_DIR / "temp"
os.makedirs(str(TEMP_DIR), exist_ok=True)
OUTPUT_FILE = TEMP_DIR / "usage_stats_kimi.json"
KEY_ID = "key7"
PROVIDER = "kimi"
def extract_metric(text, keyword, after_window=200):
"""Find percentage and date reset after a keyword in page text."""
idx = text.find(keyword)
if idx < 0:
return None, None, "unknown", f"keyword '{keyword}' not found"
after = text[idx:idx + after_window]
m_pct = re.search(r'(\d+(?:\.\d+)?)%', after)
m_date = re.search(r'(\d{4}-\d{2}-\d{2})|(\d{2}-\d{2}\s+\d{2}:\d{2})', after)
pct = float(m_pct.group(1)) if m_pct else None
date_str = (m_date.group(1) or m_date.group(2)) if m_date else None
if pct is None:
return None, None, "parse_error", f"no % found near '{keyword}'"
status = "ok"
if pct >= 95:
status = "rate-limited"
elif pct >= 80:
status = "warn"
return pct, date_str, status, None
def parse_date(dstr):
"""Parse date string to seconds remaining from now."""
if not dstr:
return 0
now = datetime.now(timezone.utc)
try:
if '-' in dstr and ':' not in dstr:
d = datetime.strptime(dstr, "%Y-%m-%d")
return max(0, int((d.replace(tzinfo=timezone.utc) - now).total_seconds()))
elif '-' in dstr and ':' in dstr:
year = now.year
d = datetime.strptime(dstr, "%m-%d %H:%M").replace(year=year, tzinfo=timezone.utc)
if d < now:
d = d.replace(year=year + 1)
return max(0, int((d - now).total_seconds()))
except Exception:
pass
return 0
def collect_metrics(page_text):
"""Parse usage metrics from page text."""
monthly_keywords = ["总使用量", "总用量", "月额度", ""]
rolling_keywords = ["5 小时用量", "5小时用量", "5小时", "5h", "5 Hour"]
weekly_keywords = ["7 天用量", "7天用量", "7天", "7d", "7 Day"]
monthly_pct = monthly_date = monthly_status = monthly_err = None
rolling_pct = rolling_date = rolling_status = rolling_err = None
weekly_pct = weekly_date = weekly_status = weekly_err = None
for kw in monthly_keywords:
monthly_pct, monthly_date, monthly_status, monthly_err = extract_metric(page_text, kw)
if monthly_pct is not None:
break
for kw in rolling_keywords:
rolling_pct, rolling_date, rolling_status, rolling_err = extract_metric(page_text, kw)
if rolling_pct is not None:
break
for kw in weekly_keywords:
weekly_pct, weekly_date, weekly_status, weekly_err = extract_metric(page_text, kw)
if weekly_pct is not None:
break
# Fallback: positional assignment if keywords miss
none_count = sum(1 for x in [monthly_pct, rolling_pct, weekly_pct] if x is None)
if none_count > 0:
print(f"[kimi] Keyword matched {3 - none_count}/3 metrics, filling rest positionally")
pct_matches = list(re.finditer(r'(\d+(?:\.\d+)?)%', page_text))
date_matches = list(re.finditer(r'(\d{4}-\d{2}-\d{2})|(\d{2}-\d{2}\s+\d{2}:\d{2})', page_text))
entries = []
for m in pct_matches:
pct_val = float(m.group(1))
pct_pos = m.start()
best_date = None
for d in date_matches:
if d.start() > pct_pos:
best_date = d.group(1) or d.group(2)
break
entries.append((pct_val, best_date, pct_pos))
metric_keys = ["monthly", "rolling", "weekly"]
current = {"monthly": monthly_pct, "rolling": rolling_pct, "weekly": weekly_pct}
for i, key in enumerate(metric_keys):
if current[key] is None and i < len(entries):
if key == "monthly":
monthly_pct = entries[i][0]
monthly_date = entries[i][1]
monthly_status = "ok"
elif key == "rolling":
rolling_pct = entries[i][0]
rolling_date = entries[i][1]
rolling_status = "ok"
elif key == "weekly":
weekly_pct = entries[i][0]
weekly_date = entries[i][1]
weekly_status = "ok"
return {
"monthly_pct": monthly_pct if monthly_pct is not None else 0,
"monthly_reset": parse_date(monthly_date),
"monthly_status": monthly_status or "ok",
"rolling_pct": rolling_pct if rolling_pct is not None else 0,
"rolling_reset": parse_date(rolling_date),
"rolling_status": rolling_status or "ok",
"weekly_pct": weekly_pct if weekly_pct is not None else 0,
"weekly_reset": parse_date(weekly_date),
"weekly_status": weekly_status or "ok",
"errors": [e for e in [monthly_err, rolling_err, weekly_err] if e],
}
def main():
print("[kimi] Starting local Playwright collection...")
with sync_playwright() as p:
# Connect to the persistent Chrome CDP service
try:
browser = p.chromium.connect_over_cdp(CDP_URL)
except Exception as e:
print(f"[kimi] ERROR: Cannot connect to CDP at {CDP_URL}: {e}")
print("[kimi] Is chrome-kimi-cdp.service running? (systemctl status chrome-kimi-cdp)")
sys.exit(1)
# Use the default browser context (has persistent profile with login session)
context = browser.contexts[0] if browser.contexts else browser.new_context()
# Check if already logged in by looking for existing kimi pages
kimi_page = None
for page in context.pages:
if "kimi.com" in page.url:
kimi_page = page
print(f"[kimi] Reusing existing Kimi tab: {page.url}")
break
if not kimi_page:
kimi_page = context.new_page()
# Navigate to the quota page
print(f"[kimi] Navigating to {KIMI_CONSOLE}")
kimi_page.goto(KIMI_CONSOLE, wait_until="networkidle", timeout=30000)
# Wait for usage data to render (poll for % signs)
page_text = ""
for attempt in range(20):
page_text = kimi_page.inner_text("body") or ""
pcts = re.findall(r'(\d+(?:\.\d+)?)%', page_text)
non_zero = sum(1 for pct in pcts if float(pct) > 0)
if non_zero >= 2:
print(f"[kimi] Page ready — {non_zero} non-zero %s (attempt {attempt + 1})")
break
time.sleep(1)
else:
print(f"[kimi] Warning: page may not have fully loaded")
# Check for login redirect
current_url = kimi_page.url
if "login" in current_url or "passport" in current_url:
print(f"[kimi] ERROR: Redirected to login page: {current_url}")
print("[kimi] Need to login first — run: python3 kimi_login.py")
browser.close()
sys.exit(1)
# Parse metrics
metrics = collect_metrics(page_text)
print(f"[kimi] Monthly: {metrics['monthly_pct']}% | "
f"5h: {metrics['rolling_pct']}% | 7d: {metrics['weekly_pct']}%")
if metrics["errors"]:
print(f"[kimi] Warnings: {'; '.join(metrics['errors'])}")
# Screenshot for debugging
try:
kimi_page.screenshot(path=str(TEMP_DIR / "kimi_console.png"))
print("[kimi] Screenshot saved")
except Exception:
pass
# Don't close the page — leave it open for next collection
# browser.close() would kill the entire CDP service
# Build output
output = {
"ok": True,
"last_refresh_iso": datetime.now(timezone.utc).isoformat(timespec="seconds"),
"provider": PROVIDER,
"accounts": [{
"key_id": KEY_ID,
"label": "key7 (Kimi)",
"provider": PROVIDER,
"workspace_id": "kimi",
"rolling": {
"usage_percent": metrics["rolling_pct"],
"reset_in_sec": metrics["rolling_reset"],
"status": metrics["rolling_status"],
},
"weekly": {
"usage_percent": metrics["weekly_pct"],
"reset_in_sec": metrics["weekly_reset"],
"status": metrics["weekly_status"],
},
"monthly": {
"usage_percent": metrics["monthly_pct"],
"reset_in_sec": metrics["monthly_reset"],
"status": metrics["monthly_status"],
},
"session_expired": False,
"last_update_iso": datetime.now(timezone.utc).isoformat(timespec="seconds"),
"error": None,
}],
}
with open(str(OUTPUT_FILE), "w", encoding="utf-8") as f:
json.dump(output, f, ensure_ascii=False, indent=2)
print(f"[kimi] Output saved to {OUTPUT_FILE}")
if __name__ == "__main__":
main()