Work · index

Projects

Two services you can buy today, one you can ask about, and the guides, tools, research and models we publish. Each entry links to its evidence.

Current offer

Service · done for you
Documents to CSV, done for you
Send up to 50 selectable-text PDFs and name the columns you need. We return a CSV in that layout, and every value cites the page and line it came from.
Try 3 PDFs free See the offer

Service · $49 per domain
Email authentication fix
We check your domain’s public SPF, DKIM and DMARC records and send the exact records your mail host needs. After you add them, we check again and show you all three at PASS. Google Workspace, Microsoft 365 or Zoho Mail.
Service · free check first
PDF accessibility remediation
We check up to five of your public service PDFs free and report every defect by document and page, then remediate: OCR, tagging, reading order, alternative text. You get the remediated files plus a PDF/UA validator report. We do not certify compliance.

Guide · redirect review
Before you replace Webflow redirects, check what disappears.
A worked, made-up comparison of kept, changed, and omitted redirect instructions — plus the checks a file review cannot make for you.

Free tool · California only
137 notices
Breach Notice Index
Search filed California breach notices for protection offers and deadline information stated in filings.
Free tool · browser check
Loadcheck
Paste a Hugging Face repo and see whether loading it would run code on your machine. Checks run in your browser.
Source

Browser games
CyberNative Games
Our games site at games.cybernative.ai. No games are live yet. The first one ships when it passes review.

Dataset · results page
94 / 100
BFCL v4 quantization pilot
We tested four GGUF quantizations on the same 100 selected BFCL V4 cases. Q8_0, Q5_K_M, and Q4_K_M each scored 94/100. Q3_K_M scored 92/100. Runs used llama.cpp's llama-server from a container image pinned by digest on the dataset card.
Dataset · reproducibility
GGUF quantization reproducibility
Two runs of one pinned quantization and evaluation harness on two small models. Quality and memory measurements matched; wall-clock load time and throughput did not.
Static snapshot
Loadcheck custom-code snapshot
A dated 200-model ranking returned by Hugging Face’s custom-code filter. It is a static reference, not the interactive check.
Security note · multi-agent
Skill directories are an instruction boundary
A per-agent skill allowlist is a control only when the skill source is per-agent. What to check when several agents share one OS user, and the fixes that bind.
Security note · PDF automation
Check that your PDF parser container is isolated
A runnable bundle that reads the docker run isolation flags back from inside a pinned parser container, with a negative control that must fail. What a pass does not show is stated up front.

Open model · provenance unverified
Colibri 8B
A Llama 3 8B chat model tuned for cybersecurity. Its training-data provenance is unverified and the model card says so — read the card before you use it.
Open model · provenance unverified
Colibri 8B · Q5 GGUF
The Q5 GGUF quantization of Colibri 8B, for running it locally. Same provenance limitation as the source model.