If you want patent protection outside the United States, yes: file before the launch. Most of the world, including Europe under Article 54 EPC, applies absolute novelty, which means anything made available to the public before your filing date, including your own launch, demo or product page, counts as prior art against you. The US is the main exception, with a one-year grace period for the inventor's own disclosures under 35 U.S.C. 102(b)(1), but relying on it forfeits Europe, China and most other markets, and it only covers disclosures that came from you.

Filing before launch does not mean finishing the patent before launch. A first filing, such as a US provisional or a European or national priority application, fixes your priority date and gives you 12 months under the Paris Convention to decide where else to file, refine the claims and add jurisdictions, while you sell openly and mark the product patent pending. The practical rule: the week the launch date goes in the calendar is the week the disclosure should go to a patent attorney. What is claimed should cover the feature that makes the launch worth announcing, not the whole product. If the launch has already happened, the US grace period may still be open; see Can I still patent an invention after publishing?

A conference talk almost always does; a pitch deck and a demo depend on who saw them and under what terms. The test for public disclosure is whether the invention was made available to at least one person who was free to pass it on. A talk, a published abstract, a poster, a recording online, a demo-day livestream or slides uploaded to a conference site are all public. A pitch to two investors who signed an NDA is not. A pitch to a room of investors who signed nothing is, in most jurisdictions, a public disclosure, even though most investors would never repeat it. Disclosure also has to be enabling: a slide saying you have a faster battery does not disclose the chemistry, but a demo that shows how it works, or a talk that explains the method, does.

Two details catch founders out. First, trade-show demos: the European Patent Convention's six-month exhibition exception (Article 55 EPC) only covers officially recognised international exhibitions, which in practice means world expos, not industry trade shows. Second, the US grace period covers your own disclosures but does nothing for Europe, where absolute novelty applies and there is no grace period. The safe sequence is to file a first application before the talk, the demo day or the pitch without an NDA, then present freely. If the disclosure has already happened, note the date, who was there and what exactly was shown; that record decides what is still patentable and where. Pasting the invention into a consumer AI assistant raises a related question: does that count as public disclosure?

In the United States, yes, for up to one year after your own publication; in Europe and most other countries, no. Under 35 U.S.C. 102(b)(1), a disclosure made by the inventor, or by someone who obtained the subject matter from the inventor, within one year before the US filing date is not prior art against the inventor. Japan, South Korea, Canada and Australia also offer 12-month grace periods, although the conditions differ and Japan's requires a declaration at filing. Europe, under Article 54 EPC, applies absolute novelty: a paper, preprint, blog post, thesis, GitHub README or conference poster published before the filing date destroys novelty for anything it enables, and the only exceptions (Article 55 EPC) are disclosures made through evident abuse against you and displays at officially recognised international exhibitions, each with a six-month window.

What this means in practice: if the paper is already out, the US route is usually still open, and a US provisional filed quickly preserves the option while you assess what else survives. Anything the publication did not enable, such as a later improvement, a specific implementation or an unpublished parameter range, remains patentable everywhere. arXiv preprints count as publications from the day they appear, so the order of operations for a research-heavy startup is: file first, post the preprint the next day. A thesis deposited in a university library is also a publication, which regularly catches spinouts. The relevant date is when the content became available, not when the journal issue printed.

You can patent the invention embodied in the code if you file within a jurisdiction's grace period, and you cannot patent it in a jurisdiction without one. Publishing source code on GitHub or under any public licence is a public disclosure of whatever the code enables. In the United States that starts the one-year clock under 35 U.S.C. 102(b)(1); in Europe it is prior art from the day the code became public, and Article 54 EPC gives no grace period. The patent covers the method or system, not the text of the code, so the question is whether the published code makes the claimed method available to a skilled person, which it usually does.

The licence you chose also matters. Apache 2.0 includes an express patent licence: every contributor grants users a perpetual, royalty-free licence to any patent claims necessarily infringed by their contribution, so you can still hold the patent but cannot assert it against users of that code. GPLv3 contains a similar patent grant in section 11. MIT and BSD say nothing express about patents, which leaves the position uncertain and is one reason enterprises are wary of them. What a patent still does for an open-source company is protect the method against competitors who reimplement it in their own proprietary code, and against uses outside the licensed contribution, which is why large open-source companies hold patent portfolios alongside their releases. If the code is not yet public, the order is the same as for a paper: file a first application, then push. Whether software is patentable at all is a separate question: see Software and AI patents.