Wind-Down Data Room: package a dead startup's code for AI-lab buyers
Self-serve tool that scrubs secrets and PII, inventories a closing startup's code and data, and drafts a sale listing. Founders pay.
Competitive · AI labs buying dead startups' archives
Written for: Solo builder
Next step
Build a CLI that runs a secrets and PII scan over a repo's full git history and outputs a one-page listing with code stats, then offer free scans to founders posting shutdown threads on Indie Hackers and r/startups.
Why now
AI-era shutdowns are piling up. Character.AI filed for Chapter 11 and is selling some assets to Disney for $20M. Most founders who are winding down don't know their repo has a price, and don't know how to strip secrets, customer PII and third-party code before selling it. A scrub-and-list tool is a weekend-sized core with a clear one-off buyer at a moment when they need money.
What became possible
Not a tech change but a new buyer. MIT Technology Review reports that AI labs are buying code and data from shut-down startups. One crypto founder netted about $25,000 for his codebase, with more pending for the data.
Three ways in
- Scrub and list CLI
Scans a git history for secrets, customer PII and vendored licensed code, rewrites the history clean, and outputs a listing with languages, LOC, commit count, test coverage and issue and PR threads, which are what labs want as training data.
- Customer
- Founders shutting down seed-stage software startups
- Model
- $299 one-off per company
- Consent checker
Reads the startup's privacy policy and ToS and flags which user data categories cannot be sold, so founders only list code and permitted data.
- Customer
- The same founders and the wind-down lawyers they hire
- Model
- Included in the paid tier
- Listing board
A curated board of cleaned archives that labs can browse, with buyers contacting sellers directly.
- Customer
- Data buyers at AI labs
- Model
- 5-10% success fee
Competition
The weak point is distribution and pricing. Most of these startups are winding down with little cash, and payouts run about $10k to $100k, so founders will not pay much upfront. The brokers already bundle cleaning into a commission. A more realistic route is to sell the scrub and consent engine to the brokers and wind-down firms, or to take a success fee, rather than charging founders directly.
Full competition notes (8 competitors)
The core opportunity is already taken. This year SimpleClosure turned selling a dead startup's code and data to AI labs into a product called Asset Hub, built into its wind-down service and reportedly behind about 100 deals. HUD and Troveo broker the same deals, and buyers such as Turing and Mercor go straight to founders. The listing board idea in particular goes head to head with SimpleClosure, which already owns the founder at the moment of shutdown, and with brokers that already have relationships with the labs. A new board would struggle with having only a few listings and few buyers at first. What is still open is the self-serve, founder-run cleaning step. The incumbents do PII scrubbing as a managed, white-glove service, and SimpleClosure keeps workspace data in beta because scrubbing is hard. No one appears to sell a CLI or tool that a founder or their lawyer can run locally. Such a tool would scan the whole git history plus the issue and PR threads for secrets, customer PII and licensed third-party code, and produce a buyer-ready report of languages, line counts, commits, test coverage and how issues link to commits. Running it on the founder's own machine means the founder never has to hand raw data to a broker. The consent checker is also unaddressed as a standalone product. Mapping privacy-policy and terms-of-service commitments to which data can be sold is exactly the legal grey area the press keeps flagging. The weak point is distribution and pricing. Most of these startups are winding down with little cash, and payouts run about $10k to $100k, so founders will not pay much upfront. The brokers already bundle cleaning into a commission. A more realistic route is to sell the scrub and consent engine to the brokers and wind-down firms, or to take a success fee, rather than charging founders directly.
- SimpleClosure Asset Hub
This is the closest match. SimpleClosure is a startup wind-down company, so it already has founders at the moment they shut down. Its Asset Hub lets them license source code to vetted buyers, and workspace data such as Slack and Jira is in beta. Forbes reported about 100 data deals in the past year, paying $10k to $100k each. SimpleClosure does the PII removal itself as a service, and calls it technically difficult.
- HUD (hud.ai)
A broker that matches startups' private repos to AI lab requests on a license-not-sale model, so the same code can be licensed to more than one buyer. It says it does light processing such as stripping PII, and claims labs pay up to $10k per qualifying codebase. It is also recruiting dead codebases on DEV Community.
- Troveo
A data-licensing broker with seller guides on selling source code and startup data to AI companies. It prices on how much production history the code has, how clean the rights are, and whether the license is exclusive.
- aitrainer.work
A seller-facing guide and funnel aimed at closed companies. It explains what labs buy (multi-contributor production code with issues, tests and deploy history) and names Turing, Mercor, DataVendor and Troveo as buyers looking for wind-down sellers.
- Turing
A data vendor that reportedly buys failed-startup codebases directly for tens of thousands of dollars, so founders can skip intermediaries.
- Mercor
Reportedly contacts shuttered or acquired startups directly (for example Warmly) to buy or license their code and task records.
- Sunset
A wind-down automation startup. Press reports say it also buys data from defunct companies at prices similar to SimpleClosure's, valuing assets by data richness, such as Jira tickets linked to commits. The reporting on this is secondhand.
- Gitleaks, git-filter-repo and BFG (open source)
Free tools, plus agent skills that wrap them, already cover secret scanning and rewriting git history. They are not aimed at preparing code for sale and do not handle PII in issue and PR threads, licensed code that was copied into the repo, or generating a listing.
Checked 2026-10-13
Risks
- Selling user data can break privacy promises and laws, so keep the product code-first
- There are few buyers and prices are opaque, and labs may stop paying as synthetic data improves
- Every customer buys exactly once, so you need a constant flow of new shutdowns
Strong tech readiness; market pull is the weak point (5).
tech 8 · pull 5 · space 7 · fit 7 · build 7
Momentum of its main topic
Sources
- AI labs are buying data from dead startups to train their modelsMIT Technology Review
- Just 5 years in, interactive chatbot platform Character.AI files for bankruptcyTechCrunch AI
- Character.AI files for Chapter 11 bankruptcy in the US, agrees to sell some assets to Disney for $20M, and plans to keep operating as it completes the sale (Rhea Rose Abraham/Reuters)Techmeme
- Character.AI files for bankruptcy, agrees to sell some assets for $20 million - ReutersReuters
What happened next
- +7d · 20 Oct 2026 nothing yet. No new product launch, funding, significant press coverage, or acquisition in the wind-down data room space was found after October 13, 2026. The field remains occupied by known competitors like SimpleClosure, Turing, and Sunset, with no evidence of a novel entrant or feature in the past week.
Status history
- 2026-10-13 Flagged as buildable