Last updated: October 2026
Magic Search is a tool to find semantically related domains in any niche in UpSnatch.
There are two modes:
AI Search: describe what domains you are looking for in plain language. An agent will make a search plan and run it for you.
AI Keyword: type a word or theme and expand it into related names with neural models (Flash Lite, Cortex, Thalamus).
What you're searching
When you search in UpSnatch, you're searching a dataset of domains that are either available or expected to become available within upcoming time windows (for example via expiring auctions, user auctions, drops, closeouts, or similar availability flows depending on the source).
UpSnatch is not a registrar and not a marketplace: you discover and evaluate domains here, and acquire them on the relevant auction/registrar platform.
Select AI Search, then describe the domains you want in one sentence: niche, listing type, quality bar, place, whatever matters.
Example:
auctions and dropped domains about personal loans for leadgen with a good backlink profile

Optionally, you can use the filters to put hard constraints on the search for the agent, like TLD, listing type, amount of backlinks, etc. But you can also just describe filters in the main input field.
The agent then runs the hunt for you:
Planner: turns your brief into search waves and keeps going until it has enough strong names.
Interpreter: asks when the brief is fuzzy, so it does not guess wrong.
Judge: weeds out irrelevant names and ranks the best matches with a Fit score.
Example: if you type bikes, that could mean brandable shop names, bike types, or anything bike-related. The Interpreter pauses and lets you choose before the hunt continues.

These are examples of filters you can write into the brief. The agent turns clear requests into hard search constraints:
| Say something like | Applied as |
|---|---|
.com and .de only / german domains | TLD filter (e.g. com, de) |
auctions / dropped domains / marketplace | Listing type |
ending this week / ending today | Auction timing window |
with backlinks / strong backlink profile | Backlink / quality floors |
DR 20+ / high trust flow / no spam | DR, TF/CF, or TF/CF ratio |
under $500 / cheap | Price filter when clear; otherwise treated as a preference |
short / brandable / no digits / no hyphens | Name shape (length, digits, hyphens) |
aged domains / at least 5 years old | Domain age |
expired forums / former local businesses | Historical site type (forum, local business, news, blog) |
Not every wish is a hard filter. If the agent would need data that we don't have available, it will tell you and ask whether to continue without it.
Metric filters need metric data. Not every domain in the catalog has every SEO metric filled in yet. If your brief asks for Trust Flow, Citation Flow, DR, or similar, the search only returns domains that have that data. Names without it are skipped for that run. You can always loosen or drop the metric ask to search a wider set.
Result cap
During a run the agent can gather up to about 2,000 candidate domains, then ranks them into the result table (capped around 1,000). If your brief is very broad, it still wraps up at that cutoff. Tighten the niche, listing type, or quality bar and run again to dig into a clearer slice.
When the run finishes, you get a finite result set: the names the agent kept for that hunt. It is not a live scroll through the whole catalog.
From there you work like usual:
Sort and filter the set (rating, spam quality, TLD, listing type, price, auction timing, and more).
Use Fit to keep the strongest brief matches near the top.
Open a domain and jump to the auction or marketplace listing when you are ready to buy.
Instant filtering
Filters apply to the finished result set instantly. You do not need to re-run AI Search just to hide weak TLDs or raise the quality bar.

AI Keyword mode finds related names for a keyword or theme. It uses neural models and AI to find semantically related domains, not only exact string matches.
To use it, select AI Keyword, type a word or short theme in the input field and pick an AI model.

There are 3 models available:
Flash Lite (quick search): fastest. Great model for a very quick keyword search.
Cortex (broad neural search): wider related and multilingual matches. Cortex neural search yields exact plus non-exact matches and reranks the results by relevance. Available from Explorer.
Thalamus (deep neural search): maximum semantic reach. This model is similar to Cortex but will also search explicitly for neurally-related non-exact matches and always adds them to the results. Available from Pro.
Going deeper
Wider models often add more loosely related names. That helps when Flash Lite feels too thin. If the niche simply has few expiring domains, deeper search mostly adds weaker associations.
You can enter:
a single keyword (e.g. dogs, pickleball, vegan), or
a phrase (e.g. dog breeds, european cruises).
Related results can include close subtopics, synonyms, alternative phrasing, and multilingual equivalents. Results still depend on what is available in the catalog at that moment.
When the search finishes, you also get a finite result set. Filter and sort it instantly (rating, spam quality, TLD, listing type, and more) without re-running the search.
Useful query shapes:
| Example query | What you get | When to use | |
|---|---|---|---|
| Broad topic | dogs | A wide mix: dog care, food, toys, adoption, etc. | Exploring, or when your niche has few expiring domains |
| Directed topic | dog breeds | Names like smallpuppies, plus breed instances such as labradors | When the broad category is too noisy |
| Few-shot examples | chihuahua, labrador, bulldog | Related instances like goldenretriever, frenchbulldog | When you want more of the same kind of name |
| Synonyms | affordable notebooks | Related wording like cheap laptops | When buyers use many different words for the same thing |
| Multilingual | dog | Equivalents like chien (French) or cane (Italian) | Always available; narrow later with TLD/country filters |
Rule of thumb:
If results feel too noisy, make the query more specific.
If results feel too narrow, broaden the query or try a wider neural model.