How long you get

Every entry in this archive has a date on it, and the date is usually the least interesting number in the file. The interesting number is the gap: how long a company gave the people using its product between the announcement and the lights going out. That interval is where a shutdown reveals what it actually thought of its users, and in the summer of 2026 the intervals got short enough to be worth writing down.

Bud gave eleven days. Figma took the team behind the prompt-to-app builder on July 7, 2026, and everything hosted on Bud and its predecessor Orchids went dark on July 18. Users were told to download their code or push it to GitHub and find somewhere else to host it. Eleven days is enough time if you are watching your inbox and your project is small. It is not enough time if you are on holiday, or if the thing you built on Bud is running something for somebody else.

Huxe gave seven. The founders pulled the audio app from both app stores in May 2026, left already-installed copies working for a week, and deleted the accounts behind them. The wind-down message was four sentences long and said the team was moving on to new things. Seven days is not a grace period so much as a courtesy, and it is worth noting that Huxe was a small company doing this with its own money, which is the situation in which a longer window costs the most.

Continue gave about four weeks, and gave them in a specific order. The acquisition notice went up on the homepage around June 16, 2026 with no press release attached. The engineers pushed a final 2.0.0 release that stripped out the telemetry and tidied the code as a deliberate handoff, then set the repository to read-only. Recurring billing was switched off, which is the detail that most distinguishes a considered wind-down from an abandoned one. On July 15 the hosted service deleted the conversation histories, saved configurations, and team settings it had been holding.

GitHub gave the most professional exit in the set, which is what you would expect from the party with the most to lose. GitHub Models closed to new customers on June 16, 2026. Full retirement was announced on July 1 for July 30. And on July 16 and July 23, GitHub deliberately broke the service for a period, so that any workflow still calling the inference API would fail loudly, on a Thursday, while somebody was awake to notice, rather than silently on the day of the shutdown. The brownout is an old operations practice and it deserves to be the norm here. It is the difference between telling people the building is coming down and testing whether they heard.

At the other end, Notion gave about three months and could afford to, because the thing at risk was never the mail. Notion Mail sat on top of Gmail; when Notion announced in June 2026 that Mail closes on September 22, the messages themselves were never in danger. What users had to rescue by hand were the drafts, the scheduled sends, the snippets, and the auto-label instructions, which is to say the work they had done teaching the product how they wanted their inbox to behave. That is the part of a productivity tool that takes months to accumulate and cannot be exported anywhere useful, and it is the part every wind-down notice underestimates.

Relay.app split the difference by wallet. New signups stopped on July 16, 2026. Free accounts end on August 15, paying customers on September 14. There is a defensible logic in giving the people with running production workflows an extra month, and there is also something clarifying about a shutdown schedule that sorts users by what they were worth. The notice opened with the words difficult news and offered no reason, which leaves the schedule as the only statement the company made about how it ranked the people it was leaving.

Read together, the intervals track something other than how much the product mattered. They track who is doing the killing. A small startup winding itself down gives days, because every additional day of servers comes out of money the founders no longer have. A platform retiring a side project gives weeks and engineers a graceful failure. A healthy company sunsetting a product line gives months, because it wants the customer to stay for the rest of the catalog. None of that is about the user. It is about the balance sheet on the other side of the announcement, and users of small products should price that in at signup rather than at shutdown.

The consistent blind spot across all of them is the same, and it is not the raw data. Most of these companies did offer an export. What almost none of them offered was a way to move the accumulated configuration somewhere it would still do anything: the auto-label rules, the saved prompts, the approval chains, the model settings, the workflow graphs. Those are the artifacts of a user's own labor, they are what make a tool feel like theirs, and they are format-locked by construction. A ZIP file of your data is not a migration path. It is a receipt.

The archive's older entries suggest this is not a new failure so much as a compressed one. Products used to die over years, with long deprecation notices and successor tools that read the old files. The generative-AI cohort dies in weeks, because it was built in weeks, on infrastructure rented by the month, by teams small enough that the wind-down is a side task for the two people not already interviewing. Speed is the defining property of this wave in both directions. The seven-day notice is not carelessness. It is what happens when the cost of keeping something alive is a monthly invoice and the cost of killing it is an email.

There is a reasonable ask buried in all of this, and it costs almost nothing. Announce the deletion date at the same time as the shutdown date, because they are not the same date and users assume they are. Turn off recurring billing on the day of the announcement, the way Continue did. Run the brownout, the way GitHub did. And publish a reason, even a thin one, because the absence of a reason does not spare anybody and it guarantees the story gets written by whoever is guessing. The obituaries here are, at bottom, the record of what companies did in the last week of a product's life. It is a small window, and it is the only part of the ending they still controlled.