ARD registry adapter for Hugging Face Skills and Spaces.
This Docker Space runs the latest published hf-discover package on Python 3.14 at
container start:
/home/user/app/start-discover.sh
The startup wrapper can run a vendored Meilisearch binary from an attached bucket and load the generated Hugging Face Skills index before starting the ARD API. If the Meilisearch bucket or skills index artifact is not mounted, the Space still starts and the combined primary search still returns Hugging Face Spaces results; indexed Skills results are added when Meilisearch is available.
Plain deployment sketch:
scripts/vendor-meilisearch.py downloads a pinned Meilisearch release binary, writes a
checksum/manifest next to it, and syncs that folder to the configured binary bucket.scripts/configure-space-runtime.py creates the configured buckets when needed, attaches
the Meilisearch binary bucket and skills-index bucket as Space volumes, sets runtime
variables, and stores MEILI_MASTER_KEY as a Space secret.start-discover.sh copies the mounted Meilisearch binary locally, verifies its
checksum from the mounted manifest, starts Meilisearch on 127.0.0.1:7700, loads the
mounted hf-skills.ndjson artifact into hf_skills, then starts the ARD API.This document is an orientation record. The scripts above are the detailed evidence and should be read directly for exact command behavior, paths, and failure handling.
Expected runtime variables:
DISCOVER_MEILI_BIN, e.g.
/mnt/meilisearch/v1.44.0/linux-amd64/meilisearchDISCOVER_MEILI_MANIFEST, e.g.
/mnt/meilisearch/v1.44.0/linux-amd64/manifest.jsonDISCOVER_SKILLS_ARTIFACT_DIR, e.g.
/mnt/skills-index/latestDISCOVER_MEILI_URL, default http://127.0.0.1:7700DISCOVER_MEILI_INDEX, default hf_skillsDISCOVER_PUBLIC_BASE_URL, optional canonical public URL used in discovery
documents, referrals, and generated skill URLs when the runtime reports an internal
base URL; leave unset/blank by defaultMEILI_MASTER_KEY, configured as a Space secretUseful endpoints:
GET /healthGET /.well-known/ai-catalog.json for ARD / AI Catalog discoveryPOST /search for combined indexed Hugging Face Skills and Hugging Face SpacesPOST /registries/huggingface/spaces/search for targeted Hugging Face Spaces searchGET /docsGET /skills/huggingface/{owner}/{space}/SKILL.mdGET /spaces/huggingface/{owner}/{space}/agents.mdRestarting or rebuilding the Space resolves the latest published hf-discover release
from PyPI.
Runtime setup helpers from the repo root:
uv run scripts/vendor-meilisearch.py
uv run scripts/configure-space-runtime.py --set-secret --generate-secret
3 commits
ARD registry adapter for Hugging Face Skills and Spaces.
This Docker Space runs the latest published hf-discover package on Python 3.14 at
container start:
/home/user/app/start-discover.sh
The startup wrapper can run a vendored Meilisearch binary from an attached bucket and load the generated Hugging Face Skills index before starting the ARD API. If the Meilisearch bucket or skills index artifact is not mounted, the Space still starts and the combined primary search still returns Hugging Face Spaces results; indexed Skills results are added when Meilisearch is available.
Plain deployment sketch:
scripts/vendor-meilisearch.py downloads a pinned Meilisearch release binary, writes a
checksum/manifest next to it, and syncs that folder to the configured binary bucket.scripts/configure-space-runtime.py creates the configured buckets when needed, attaches
the Meilisearch binary bucket and skills-index bucket as Space volumes, sets runtime
variables, and stores MEILI_MASTER_KEY as a Space secret.start-discover.sh copies the mounted Meilisearch binary locally, verifies its
checksum from the mounted manifest, starts Meilisearch on 127.0.0.1:7700, loads the
mounted hf-skills.ndjson artifact into hf_skills, then starts the ARD API.This document is an orientation record. The scripts above are the detailed evidence and should be read directly for exact command behavior, paths, and failure handling.
Expected runtime variables:
DISCOVER_MEILI_BIN, e.g.
/mnt/meilisearch/v1.44.0/linux-amd64/meilisearchDISCOVER_MEILI_MANIFEST, e.g.
/mnt/meilisearch/v1.44.0/linux-amd64/manifest.jsonDISCOVER_SKILLS_ARTIFACT_DIR, e.g.
/mnt/skills-index/latestDISCOVER_MEILI_URL, default http://127.0.0.1:7700DISCOVER_MEILI_INDEX, default hf_skillsDISCOVER_PUBLIC_BASE_URL, optional canonical public URL used in discovery
documents, referrals, and generated skill URLs when the runtime reports an internal
base URL; leave unset/blank by defaultMEILI_MASTER_KEY, configured as a Space secretUseful endpoints:
GET /healthGET /.well-known/ai-catalog.json for ARD / AI Catalog discoveryPOST /search for combined indexed Hugging Face Skills and Hugging Face SpacesPOST /registries/huggingface/spaces/search for targeted Hugging Face Spaces searchGET /docsGET /skills/huggingface/{owner}/{space}/SKILL.mdGET /spaces/huggingface/{owner}/{space}/agents.mdRestarting or rebuilding the Space resolves the latest published hf-discover release
from PyPI.
Runtime setup helpers from the repo root:
uv run scripts/vendor-meilisearch.py
uv run scripts/configure-space-runtime.py --set-secret --generate-secret
3 commits