AI tool lock-in: the five things to check before you commit
Lock-in in AI tools comes from five places: gated export, a hosted backend your exported code still calls, non-portable trained assets, plan-bound commercial licences, and proprietary formats. Check all five during a trial, because each one is cheap to test and expensive to discover later.
By the AI Alt Finder team
Lock-in is rarely a clause in a contract. It is an accumulation of small design decisions that each look reasonable and together mean leaving costs more than staying. All five below can be tested during a free trial in under an hour, which is the only time testing them is free.
1. Is export gated behind a higher tier?
The most common form. Generation is available on the entry plan, getting your work out is not. Base44's documentation, for example, states that GitHub two-way sync requires its Builder plan or higher (docs.base44.com, checked 28 July 2026), while Lovable documents two-way sync with github.com on all plans (docs.lovable.dev, same date).
Same category of product, opposite answers to the same question. Price the tier that includes export, because that is the real entry price for anyone building something they intend to keep.
2. Does the exported artefact run without the vendor?
Export can be technically complete and practically useless. Where a platform generates a client that calls its own hosted backend, cloning the repository gives you code that will not function until you replace that backend.
The test takes ten minutes: clone it, remove the vendor's environment variables, and try to start it. Whatever fails is the part you do not own.
3. Can trained or cloned assets leave?
Voice clones, trained styles and fine-tunes are the least portable things you will create, and often the most valuable. Assume none of them transfer, then keep the source material they were built from so the asset is rebuildable elsewhere.
This is the item people discover on the day they leave, which is exactly the wrong day.
4. Is your licence bound to the plan?
On several platforms the right to use output commercially is a plan feature. ElevenLabs, for instance, lists a free tier with 10,000 credits per month, with the commercial licence starting at its first paid tier (elevenlabs.io/pricing, checked 28 July 2026).
Downgrading a plan can therefore change the legal status of work you already published. Check whether the licence covers output created while subscribed or only while you remain subscribed, because those are meaningfully different products.
5. Are open weights the same as an open licence?
No, and the gap is expensive. Black Forest Labs distributes FLUX.1 [schnell] under Apache-2.0 while FLUX.1 [dev] carries a non-commercial licence requiring a separate paid licence for commercial activity (bfl.ai licence terms and the model's LICENSE file, checked 28 July 2026).
Being able to download and run a model says nothing about being able to sell what it makes. Read the licence file, not the download page.
Lock-in is a price, not a disqualification
Every managed product trades some portability for convenience, and that is frequently a good trade. The mistake is making it accidentally. Decide what leaving would cost while you are still choosing, write the number down, and revisit it when the renewal arrives.