Skip to content

About

Free, private, 100% client-side log file analyzer: GPTBot vs OAI-SearchBot cost split, crawl budget, Cloudflare/Fastly/AWS edge rules + robots.txt generator. Training vs search-index vs user-triggered AI split (2026).

Topics

Resources

Contributing

Security policy

Stars

1 star

Watchers

0 watching

Forks

Latest commit

 

History

49 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Log File Bot Traffic Cost Analyzer

Live Demo 1.02GB Proven Bot DB Tests Price MIT CI

Drop 1M-line logs → see GPTBot vs OAI-SearchBot cost split → copy the Cloudflare rule. Free, private, $0 vs $139/yr Screaming Frog Log Analyser ($279 Spider) / €171–383/mo JetOctopus (prices verified Feb 2026). No log leaves your machine.

Try Live · 1-click 10k demo · 1GB proof teardown

Enterprise UI (v2.3.0): Material 3 (Material You) theme — top app bar with Analyzer · Knowledge Base ▾ · About · How To Use (guides live on their own pages, not in a sidebar), M3 filled/tonal pill buttons, tonal chips, elevated cards, and linear progress. Centered 1200px layout with no right-side whitespace gap, tables scroll inside their cards, and a crisp SVG brand mark. Sample downloads by URL range with live max-size estimate (0–100 … 1M–10M URLs; in-browser generation caps at 1M, CLI covers the rest), and the 2026 Pixel typeface (Google Sans Flex) throughout.

Private by design: 100% client-side, no XHR/WebSocket/sendBeacon. Only network request is loading the demo file you click. See METHOD.md.

Drop → cost split → Cloudflare rule in 30s KPI strip from the measured 1.02GB run

1. Proven on a real 1.02 GB / 3.2M-line log

Headless-Chromium run, 2026-09-11, oneGB.jsonl (seed 7). Systematic 1-in-10 sample, honestly bannered.

Measurement Result
Read + parsed + analyzed 320,000 of 3,200,000 lines in 6.6 s, 13.0 s wall total
Records / bandwidth (sample) 320,000 records, 10.92 GB, 223,878 unique IPs
Traffic split 70.0% human, 15 categories detected
Cost, CloudFront defaults $2.61 total, $0.25 blockable (training + suspicious only)
Response profile avg 799 ms, P95 1,474 ms, P99 1,535 ms
Threats / verification flags 19,272 suspicious, 427 unusual checks surfaced
Crawl-budget waste 18% trap bandwidth quantified

Full numbers + 29 screenshots: research/1gb-teardown.html. Fixture: npm run gen-logs -- --seed 42 --days 30 reproduces sample-data/EXPECTED.md byte-identically.

2. Why this exists (2026 position)

Tool Price 2026 Scale Privacy When to use what
This tool $0 Browser <500MB triage, CLI 500MB–50GB exact, BQ 50GB+ 100% client-side 30-sec answer, DPDP/RBI logs that can't leave the machine
Screaming Frog LFA $139/yr (Spider $279/yr) Desktop, slows past a few M lines Local desktop Mature docs, Spider × log joins
JetOctopus €171–383/mo Cloud millions, logs+crawl+GSC joined Cloud upload Continuous monitoring, history, client dashboards
OnCrawl / Botify €49–2,500/mo Up to 500M lines/day Cloud + FTP Enterprise SSO/procurement
Splunk Custom ingest Infinite Cloud Build-it-yourself DIY

Honest take: paid tools win on continuity/history. This wins on privacy + speed + price. Details: Frog vs JetOctopus vs $0.

3. The 2026 distinction: training vs search-index vs user-triggered

Bot DB v2026.09.17, 127 signatures. Interviewers ask exactly this.

Class Examples Edge action
Training GPTBot, ClaudeBot, CCBot, Bytespider Block/throttle freely, 60/min (20 aggressive)
Search-index OAI-SearchBot, PerplexityBot, Claude-SearchBot ALLOW 120/min, 429 + Retry-After only — never hard block
User-triggered ChatGPT-User, Perplexity-User, OAI-AdsBot (allow-listed) Do NOT throttle (300/min abuse ceiling only)

Includes Cloudflare Sept-15-2026 defaults, Pay Per Crawl 402 beta, Web Bot Auth, vendor IP JSON (data/bot-ips.json, 14 sources, full CIDRs + IPv6). See BOTS.md.

4. The 10 modules (live demo, screenshots from 1GB run)

# Module What you get
1 Bot Classification 127 sigs + 18 browser patterns, bot-first order, bandwidth/IPs/TTFB per category
2 Verification (4 layers) Vendor IP JSON → TLS → cloud-ASN heuristic → behavior; UNVERIFIED spoof KPI, opt-in DoH rDNS
3 Crawl Budget + Render Gap 16 traps (gclid, cart, /_next/data/) + JS-shell <5KB hunt (zero GEO citation chance)
4 Cost Measured bytes × CDN preset, 95% interval, blockable = training + suspicious only
6 Edge Rules + Policy Cloudflare/Fastly/AWS copy-paste, robots + llms.txt + crawlers.json bundle, OAI-AdsBot guard
7–8 Performance + Traffic TTFB bots vs humans, bursts (z-score, no deps), 404-cluster tickets, GSC overlay
9 Security Traversal, .git/HEAD, creds, SSRF, velocity anomalies
10 CFO / FinOps Totals, monthly/annual run-rate, CSV + redesigned 2-page PDF (M3 blue, numbered sections, KPI cards, dynamic exec box, aligned tables + sign-off) — vendored jsPDF, no CDN. Full Report PDF ↓ exports all 10 modules + CFO summary in one offline reference file

Screenshots: assets/screenshots/09-tab1-classification.png → 29-cfo-viewport.png (analysis modules unchanged; v2.2.1 refreshed the app shell centering, brand mark, and CFO PDF design). Full history: CHANGELOG.md.

5. Quick start

git clone https://github.com/dipakjad1993/Log-File-Bot-Traffic-Cost-Analyzer.git
cd Log-File-Bot-Traffic-Cost-Analyzer
node dev-server.js   # LOCAL DEV ONLY — http://localhost:8080. Never deploy to Pages.
# or: docker build -t log-analyzer . && docker run -p 8080:8080 log-analyzer
npm test             # 121 asserts
npm run lint

Inputs: JSON array, JSONL/NDJSON, Apache Combined, Nginx, W3C Extended (#Fields), AWS ALB, Cloudflare Logpush, .gz — mixed files + rotations OK. Streams in 8MB slices; 500MB+ exact → node tools/cli.js --exact --in access.log --out summary.json. Test data: pick a URL range in the tool (0–100 / 100–1K / 1K–10K / 10K–100K / 100K–1M / 1M–10M, each showing its max file size, ≈450 bytes/row; browser builds up to 1M) or npm run gen-logs -- --lines N --out file.jsonl.

Deploy: Cloudflare Pages (primary, main, no build command, headers from _headers) or GitHub Pages backup. Live: https://log-file-bot-traffic-cost-analyzer.pages.dev

6. Method + honest limits

  • Cost = measured bytes × pricing. Blockable = training + suspicious only (asserted in tests). Origin-compute OFF by default.
  • Verification = vendor IP JSON first, heuristics labeled low-confidence, never verdicts. Sampling disclosed with stride/counts/timing.
  • Limits: Browser <500MB triage (300k sampled) · CLI 500MB–50GB exact · BigQuery 50GB+. See METHOD.md.
flowchart LR
  Drop[Drop logs .gz/multi-file] --> Slice[8MB slices + auto-detect]
  Slice --> Worker[Web Worker analyze 14 steps]
  Worker --> Render[10 renderers: bots/verify/cost/edge/CFO]
  Render --> Export[CFO PDF + CSV + Cloudflare rules + llms.txt]
  Worker --> CLI[CLI --exact 500MB-50GB]
  CLI --> BQ[BigQuery pack 50GB+]
Loading

Docs

BOTS.md · METHOD.md · ROADMAP.md · BIGQUERY.md · sitemap.xml

Contributing / Security / License

CONTRIBUTING.md · SECURITY.md · MIT (LICENSE) — PRs welcome (good first issue labeled). If this saved you €383/mo, please star the repo.

About

Free, private, 100% client-side log file analyzer: GPTBot vs OAI-SearchBot cost split, crawl budget, Cloudflare/Fastly/AWS edge rules + robots.txt generator. Training vs search-index vs user-triggered AI split (2026).

Topics

Resources

Contributing

Security policy

Stars

1 star

Watchers

0 watching

Forks

Contributors

Languages