robots.txt
See which AI crawlers other sites block →# As a condition of accessing this website, you agree to abide by the following
# content signals:
# (a) If a Content-Signal = yes, you may collect content for the corresponding
# use.
# (b) If a Content-Signal = no, you may not collect content for the
# corresponding use.
# (c) If the website operator does not include a Content-Signal for a
# corresponding use, the website operator neither grants nor restricts
# permission via Content-Signal with respect to the corresponding use.
# The content signals and their meanings are:
# search: building a search index and providing search results (e.g., returning
# hyperlinks and short excerpts from your website's contents). Search does not
# include providing AI-generated search summaries.
# ai-input: inputting content into one or more AI models (e.g., retrieval
# augmented generation, grounding, or other real-time taking of content for
# generative AI search answers).
# ai-train: training or fine-tuning AI models.
# use: how AI systems may consume the content (immediate, reference, or full).
# ANY RESTRICTIONS EXPRESSED VIA CONTENT SIGNALS ARE EXPRESS RESERVATIONS OF
# RIGHTS UNDER ARTICLE 4 OF THE EUROPEAN UNION DIRECTIVE 2019/790 ON COPYRIGHT
# AND RELATED RIGHTS IN THE DIGITAL SINGLE MARKET.
# BEGIN Cloudflare Managed content
User-agent: *
Content-Signal: search=yes,ai-train=no,use=reference
Allow: /
# Training crawlers
User-agent: Amazonbot
Disallow: /
User-agent: Applebot-Extended
Disallow: /
User-agent: Bytespider
Disallow: /
User-agent: CCBot
Disallow: /
User-agent: ClaudeBot
Disallow: /
User-agent: Diffbot
Disallow: /
User-agent: Google-Extended
Disallow: /
User-agent: GPTBot
Disallow: /
User-agent: omgili
Disallow: /
User-agent: anthropic-ai
Disallow: /
User-agent: Claude-Web
Disallow: /
User-agent: cohere-ai
Disallow: /
User-agent: MistralAI-Training
Disallow: /
User-agent: meta-externalagent
Disallow: /
User-agent: GoogleOther
Disallow: /
User-agent: Baiduspider
Disallow: /
User-agent: PetalBot
Disallow: /
User-agent: AwarioSmartBot
User-agent: AwarioRssBot
Disallow: /
User-agent: Google-CloudVertexBot
Disallow: /
User-agent: QualifiedBot
Disallow: /
User-agent: Cotoyogi
Disallow: /
User-agent: ICC-Crawler
Disallow: /
User-agent: atlassian-bot
Disallow: /
User-agent: FishBot
Disallow: /
User-agent: BorderxBot
Disallow: /
User-agent: NavuBot
Disallow: /
User-agent: SemrushBot-SWA
Disallow: /
User-agent: WARDBot
Disallow: /
User-agent: magpie-crawler
Disallow: /
User-agent: KimiBot
Disallow: /
User-agent: CitibotSiteCrawler
Disallow: /
# END Cloudflare Managed Content
# Retrieval agents — cite content in live search responses, allow full access
User-agent: OAI-SearchBot
Allow: /
User-agent: Claude-Web
Allow: /
User-agent: PerplexityBot
Allow: /
User-agent: Google-Extended
Allow: /
User-agent: Applebot-Extended
Allow: /
# Training scrapers — consume bandwidth with no referral traffic, block
User-agent: GPTBot
Disallow: /
User-agent: ClaudeBot
Disallow: /
User-agent: CCBot
Disallow: /
# All other crawlers
User-agent: *
Allow: /
Sitemap: https://hitl.systems/sitemap.xml
LLMs-txt: https://hitl.systems/llms.txt
llms.txt
Compare with other sites' llms.txt files →# HITL Systems
> HITL Systems builds AI resolution infrastructure: open-weight models that run on-device, requests that escalate to the cloud when needed, and human resolution when models aren't enough.
## Sections
- [Articles](https://hitl.systems/articles/): Articles and guides.
- [Books](https://hitl.systems/books/): Long-form guides organised into chapters.
## Introducing latai: AI Help for the Files That Cannot Leave Your Desk
latai is a 22 MB on-device AI runtime for the files you are not allowed to upload. It runs small signed models on your own computer or phone, so the document, the question and the answer never leave it. One call, six SDKs, no server, no cost per answer.
https://hitl.systems/articles/introducing-latai-on-device-ai-runtime/
## The AI Innovation Paradox Is Real. Your Evidence For It Probably Isn't.
Almost 90% of companies use AI and 6% see profit from it. We checked the studies behind the AI innovation paradox and the tools that claim to fix it. The strongest finding is the one nobody quotes.
https://hitl.systems/articles/ai-innovation-paradox/
## Introducing maap, a System One Estimation Model
maap reads a record as text and returns a distribution for the number you asked for: a median, your quantiles, an interval fitted to held-out data, and the probability of crossing a threshold. 4B parameters, one forward pass.
https://hitl.systems/articles/introducing-maap-estimation-model/
## What Did Not Work: 48 Failed Ideas From an Open LLM Classifier
An open LLM classifier project logged 48 failed ideas with the numbers attached. Four assumptions broke, one lever worked, and the ledger can save you months.
https://hitl.systems/articles/llm-classifier-negative-results/
## vLLM vs SGLang: How to Serve Your Own LLM Without Paying for Empty GPU Memory
Size GPU memory per user first, pick vLLM or SGLang with one question, then test both on your own prompts before you buy hardware. Sources and math included.
https://hitl.systems/articles/vllm-vs-sglang-serve-your-own-llm/
## How to Make Money With AI in 2026: Four Doors to Check Before You Build
AI made building cheap for everyone. Retention data, Tesla's presale, and a four-question test to check an AI idea gets paid before you build it.
https://hitl.systems/articles/how-to-make-money-with-ai-2026/
## ChatGPT Dots Security Audit: The Slack Second Door and 6 More Gaps to Test
OpenAI's docs say Slack channel limits don't block DMs and dot memories can't be edited. Seven tests to run before a small-team ChatGPT rollout.
https://hitl.systems/articles/chatgpt-dots-rollout-audit/
## How to Write With AI Without Sounding Like AI: A Stage-Gated System
Why AI writing sounds robotic (two studies), the tells to lint for, and a stage-gated system that keeps the speed and drops the slop.
https://hitl.systems/articles/how-to-write-with-ai-without-sounding-like-ai/
## OpenAI DevDay 2026, Audited: Dots, GPT-6.1 Sol, and the Plan That Got Halved
DevDay 2026 audited, not recapped: the Pro 200 halving math, the dot usage fine print, GPT-6.1 Sol's $5.47-per-task ledger, and safety claims vs the record.
https://hitl.systems/articles/openai-devday-2026-audit/
## Jev Open Source Tools, Sorted: What's Worth Installing
300+ Jev repos, no triage. The shortlist with verdicts, install configs, per-tool costs, and the claims the hype cycle got wrong.
https://hitl.systems/articles/jev-open-source-tools-audit/
## Opus 5.5 vs Grok 4.7 vs GPT-6: Who's Actually Cheaper Per Task?
Sticker prices say one thing, invoices another. The effort-level cost swing, token math, and fine print for Opus 5.5, Grok 4.7, and GPT-6, audited.
https://hitl.systems/articles/opus-5-5-grok-4-7-gpt-6-cost-audit/
## GPT-6 Sol, Audited: The 50% Price Cut vs the Benchmarks Nobody Cited
GPT-6 Sol, audited: the 50% price cut next to OpenAI's own prior benchmarks, GDPval and DeepSWE regressions, the 272K rate trap, and per-task cost math.
https://hitl.systems/articles/gpt-6-sol-audit/
## Jev Can't Hallucinate. We Audited What It Does Instead
TypeSafe admits the 0% is definitional, not measured. What Jev's hallucination claim hides: wrong-but-valid picks, calibration data, Jev vs Laya numbers.
https://hitl.systems/articles/jev-hallucination-audit/
## Antigravity CLI, Audited: Speed Claims, Rate Limits, and Real Migration Cost
Antigravity CLI, audited: speed claims vs tests, rate limits by tier, per-task quota math, migration failures, and the closed-source trade.
https://hitl.systems/articles/antigravity-cli-audit/
## ChatGPT Voice Mode, Audited: Limits, Failures, and Real Cost
ChatGPT Voice Mode, audited: tiers, usage limits, failure modes, privacy, and the real API cost math. Every claim labeled verified, reported, or derived.
https://hitl.systems/articles/chatgpt-voice-mode-audit/
## Agent-Native Apps: A Builder's Guide to BYOA, MCP, and Not Breaking Your Product
Users bring their own agents now. A builder's guide to agent-native apps: BYOA, MCP vs CLI vs API, how it breaks, and how to price it.
https://hitl.systems/articles/agent-native-apps-byoa-guide/
## What Is Jev? The AI Decision Model, Audited by an Engineer
Jev can't write a word. Is it really 200x faster and 400x cheaper? An engineer's audit of the claims, the pipeline math, and the failure modes.
https://hitl.systems/articles/what-is-jev-ai-model/
## Contact
Email: hello@hitl.systems