iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
OriginTrace is a prototype that searches for possible copies of DEV.to articles, compares their text, and checks for visible attribution. It can help organize leads, but it cannot establish that a page is stolen: its author reports no benchmark or precision-and-recall results, and the system’s labels are not legal findings.
What OriginTrace does
Dhruv Jani describes OriginTrace as an AI-assisted content-provenance workflow for DEV.to authors trying to distinguish unauthorized copying from legitimate syndication. A user supplies a DEV.to article URL. The system extracts distinctive phrases, searches the web through Serper.dev, gathers up to 15 candidate pages, compares their text, checks for attribution, and stores the results as structured data in Sanity. An AI agent can then answer questions about those records or draft a takedown notice. This is the project author’s description, not an independent verification of the implementation or its results. OriginTrace project description, October 4, 2026.
How it finds and compares candidate copies
Search and candidate collection
The described process starts by extracting distinctive phrases from the submitted DEV.to post and searching the web for them. It then scrapes up to 15 candidate pages for comparison. This approach can surface pages that share searchable wording, but it depends on what the search returns and what the scraper can retrieve; the article does not report recall across the web or coverage of all reposts.
Three text-overlap signals
For each candidate, OriginTrace combines three measures. The project article gives these implementation weights:
#1 Best Overall
| Signal | Weight in the described score | What it indicates |
|---|---|---|
| Word overlap (Jaccard similarity) | 20% | How much the two texts share words. |
| 5-gram shingling | 50% | How much they share sequences of five words. |
| Longest common subsequence (LCS) | 30% | How much ordered text the two documents share, even when parts differ. |
These percentages describe how the project weights its signals; they are not accuracy rates or evidence that a particular score proves copying. The project article provides no precision, recall, false-positive rate, or evaluation dataset.
How attribution affects the result
Text similarity and attribution are separate checks. The project says it looks for the author’s name, a link to the original DEV.to post, or wording such as “via” and “originally published on.” It labels results as credited_syndication or unattributed_repost. A high-overlap page with a clear link back may therefore be treated differently from one without visible attribution.
That distinction is useful for review, but absence of a detected attribution signal does not prove a page is infringing. The labels describe OriginTrace’s classification, not a legal conclusion. Jani also said the parser checks visible links in the page body and does not yet inspect the HTML head’s canonical URL, so a canonical tag may not be reflected in the attribution check. Jani’s discussion response on canonical metadata, October 5, 2026.
Recommended Free Tools
What the Sanity Context MCP and AI agent contribute
The project describes Sanity as a structured ledger for checks, with fields such as numeric overlapPercent, a verdict, attribution booleans including hasAuthorName and hasOriginalLink, and arrays for signals or missing evidence. Its agent connects to Sanity Context MCP over HTTP, requests initial context, and uses groq_query to query the Content Lake. A backend fetchSanityContext() function also prefetches an aggregate snapshot before the agent reasons.
In practical terms, this lets the agent answer questions that combine stored fields—for example, which candidate had the highest overlap, whether that candidate credited the author, or which results lack an original link. The project author presents this as an architecture for querying structured records; the article does not establish that it outperforms other search approaches.
The listed implementation stack includes Next.js 16, NVIDIA NIM and Nemotron 3.5 Lightning, the Vercel AI SDK and @ai-sdk/mcp, Sanity Content Lake and Context MCP, Serper.dev, a custom TypeScript overlap engine, Mozilla Readability, Cheerio, JSDOM, and Render. These are components named in the project description, not a recommendation or confirmation of their current versions, pricing, or availability.
What OriginTrace can—and cannot—establish
The most important limitation is validation. In a follow-up discussion, Jani said the hackathon project did not have a formal hold-out set of known syndications and stolen posts for calculating precision or recall. No independent performance statistic is reported, so there is no substantiated accuracy figure for deciding whether a match is theft or credited syndication. Jani’s discussion response on evaluation and human review, October 6, 2026.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Use a result as a lead to inspect, not as a conclusion. Confirm that the candidate actually reproduces your work, review its attribution and publication context, and preserve relevant evidence before deciding how to respond. A similarity score alone cannot determine authorization, ownership, or infringement.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.DMCA drafts require human review
The project says its agent can draft a DMCA takedown notice, but does not send it automatically; the author retains the final review step. Treat any generated text as a draft: verify the page, the original work, the facts in the notice, and the appropriate recipient before taking action. An automated match is not a substitute for making sure a complaint is accurate.
Scope and maturity
Jani frames OriginTrace as a working hackathon-scope prototype, with the AI agent and Sanity ledger in beta. The described platform scope is DEV.to. Authentication and per-author data isolation, support for platforms beyond DEV.to, scheduled monitoring, and embedding-based paraphrase detection are listed as future enhancements, not established current capabilities.
The project article describes a globally queryable public data ledger and publishes a Sanity project ID and dataset. That is the author’s description; it is not an independent privacy or security audit. Ideas discussed for signed identity and purpose-tagged access logs were proposed future architecture, not features the project says it already has.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
