Searching the registry

Find models across every namespace you can see.

Search models#

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19

from scaiatlas import ScaiAtlas
client = ScaiAtlas.from_env()

# Simple query
for m in client.search.models("llama"):
    print(m.full_name, m.type)

# With filters
results = client.search.models(
    "instruct",
    type="llm",                 # a model type; see the reference
    tag="llama",                # models carrying this tag
    namespace="poolnoodle",     # restrict to one namespace
    sort="downloads",           # updated | created | name | downloads
    order="desc",               # asc | desc
)
print(results.pagination.total_count, "matches")
for m in results:
    print(m.full_name, m.download_count)

Iterate every page automatically:

1
2

for m in client.search.models_all("embedding", type="embedding"):
    print(m.full_name)

Search namespaces#

1
2

for ns in client.search.namespaces("pool"):
    print(ns.name, ns.visibility)

What you can see#

Search results are scoped to what you're allowed to read: public namespaces plus those in your tenant (and any you're an explicit member of). You'll never see models in namespaces you can't access.

Notes#

• Query text is matched against model names, display names, descriptions, and tags.
• sort controls ordering; combine with pagination (page, per_page).
• Results validate into the same Model objects returned elsewhere, so m.full_name, m.type, m.download_count, etc. are all available.

Under the hood (REST)#

• GET /api/v1/search/models — query params q, type, namespace, tag, sort, plus pagination.
• GET /api/v1/search/namespaces — query param q plus pagination.

See the REST API reference.