How LanguageOps handles your text, what is verified, and where the boundaries lie.
LanguageOps does not attempt to prove whether text was written by a human or a machine. It does not output probability percentages or run style matching. Instead, it scans for formulaic, AI-associated writing patterns (like "delve", "tapestry", or repetitive sentence rhythms) that weaken communication.
All anti-pattern scanning is executed locally on our gateway or extension using deterministic regular expressions and heuristics. Your original drafts are checked for style markers without calling external APIs during the scanning phase.
If you use hosted LanguageOps, submitted drafts pass through the LanguageOps server and may be sent to the configured LLM provider for refinement. Provider keys are used only for the request and are not stored or logged. If you self-host the gateway, that processing happens on your own infrastructure.
To prevent AI hallucinations or loss of information, LanguageOps runs automated preservation checks. These checks scan and align critical factual anchors before and after refinement, including:
While the verification engine significantly reduces the risk of incorrect rewrites, preservation checks are heuristics and do not eliminate all risk of semantic change or factual omission. High-stakes communication, documentation, and critical professional summaries must still be reviewed by a human before sending.