Legwork, a Windows OSINT AI investigator that cites every sentence
A two-day-old GitHub project called Legwork answers plain-language questions from public pages and attaches a source to each line it writes.
A citation on every sentence
A GitHub repository named StormBrawlerSack/legwork-osint-ai-investigator is drawing attention for a narrow promise: an answer where every sentence carries the public pages it came from. As of 2026-11-15 the project had 326 stars and 26 forks, all gathered in the two days since it was created on 2026-11-13. It is written in JavaScript and released under the MIT license. The repository describes the app itself as Legwork, an OSINT AI investigator for Windows, currently at version v1.2.1 and available as a download. The full write-up lives on the project's GitHub page.
What the source confirms
The source states the basic shape of the product. You ask a question in plain words, Legwork reads the public pages it finds, and it writes an answer with a citation attached to each sentence. The repository lists topics including ai-agent, ai-investigator, citations, investigation, llm, osint, research-assistant and osint-tool, which places it in the research-assistant corner of the agent boom rather than the coding-tool corner.
The stated motivation is familiar to anyone who has done desk research: twenty tabs, a notepad, and a question that keeps growing. The source frames the difference as questions versus queries. A search engine returns a list of results. Legwork, in the source's description, reads those results and produces a written answer.
The rules it claims to follow
The source describes three constraints that make the output checkable. First, no source means no sentence: every line of an answer carries the pages behind it. Second, claims resting on a single source are left out. Third, gaps in the evidence are stated rather than filled in. The app also reports one number alongside the answer: how many public pages it read to produce it.
The demo described on the page shows the same idea in motion. A question types itself in, a page counter runs up, and the answer appears line by line. Key phrases are marked with a highlighter and each sentence carries numbered source chips, with source cards listed below the answer. The design is deliberately paper-like: ink #121114, paper #F4F1EC and a highlighter pink #FF85C0 for anything sourced. Three sample questions can be switched beneath the answer, and the line that actually answers the question is held back so the download card takes its place.
What it changes for people building apps
The useful part here is not the Windows packaging, it is the citation rule as a product decision. Dropping single-source claims and printing the gap where the evidence runs out is a hard constraint to design around, and it is the kind of constraint that makes an AI feature auditable rather than merely fluent. Anyone adding a research or summarising feature to an app faces the same trade: a confident paragraph with no traceable origin, or a shorter answer with links a user can click. Legwork's choice is the second one, and the page count gives the reader a rough sense of how much work went into the reply.
Limits and open questions
The source does not say which language model Legwork runs on, how it discovers the pages it reads, whether it works offline, what the download costs, or whether the app can be built from source despite the MIT license. It also does not describe accuracy on contested questions, only the rule about single-source claims. Windows is the only platform mentioned. Star counts in the low hundreds over two days measure attention, not correctness, and the repository is two days old, so there is no track record to check against the claims on the page.
For builders adding AI to an app
The grounding pattern is the transferable idea: constrain the model to what it can cite, and show the sources next to the sentence they support. If you are wiring a model into an app you are building, Draftship's documentation on adding AI to your app covers how that step fits into a project. Draftship is not affiliated with Legwork and does not support or integrate with it.
Primary sources
- StormBrawlerSack/legwork-osint-ai-investigator: AI OSINT investigator for Windows: ask a question in plain words and get an answer built from public sources, with a source on every sentence. · GitHub trending AI repositories, Nov 15, 2026
This article was drafted with AI assistance from the primary sources listed below and checked automatically against them. An editor hasn't reviewed it yet.