Questions, answered
What is Tethered Trust?
A compatibility decision service for shopping agents and the people they buy for. Send what someone owns and what's in the cart; get back one decision (supported, conditional, incompatible or insufficient evidence), the evidence behind it, the questions still blocking it, and a signed receipt.
Who is it for?
Developers building shopping, PC-configuration, repair and replacement-part agents. The website checker is free for anyone.
How is this different from asking an AI model?
Answers come from manufacturer-sourced records and fixed rules, not from a model's memory. Every answer links its sources and says when it can't decide, instead of guessing. AI is only used to read specs for products we don't know yet, and those readings are marked and held for review.
What does it cover?
PC building (CPU sockets, cooler mounts, memory), laptops and consoles (RAM, M.2 SSDs, USB-C power), phones and watches, cameras, cordless power tools, printers, refrigerator filters, vacuums, air purifiers and HVAC media, e-bike batteries and smart-home hubs. See the coverage page for every model.
How accurate is it?
On our 50-question benchmark, sourced from manufacturer pages, it has given zero wrong “it fits” answers. The full results, including the questions we couldn't answer, are on the benchmark page.
What does a receipt prove?
That Tethered Trust issued that exact decision at that time and nobody has changed it since. It doesn't prove the decision was right. Recheck a plan before ordering to catch changes since it was issued.
How much does it cost?
Free during early access within rate limits. Paid endpoints cost $0.05 per completed check or $0.25 per completed purchase plan, paid with prepaid credits. You're never charged for errors or “need more detail” answers, and the price never depends on the verdict.
Is there a sandbox?
Yes. Every endpoint has a sandbox twin under /v1/sandbox: same engine and catalog, free, unmetered (300 calls an hour), never charged. Documented scenarios with fixed answers are at /v1/sandbox/scenarios.
What happens when you don't know a product?
You get insufficient_evidence with the exact question to ask. The model goes into our research queue, ranked by how many integrations asked for it.