Store

Browse approved apps, skills, agents, and flows from the public store.

These endpoints power Apps and the skill registry. They return store listings (metadata, category, rank, tags) — not full app manifests or skill instructions.

No API key is required. See REST overview — response format.

→ Search API — semantic discovery across apps, skills, and knowledge
→ Skills API — skill content, versions, and management
→ Apps API: your workspace's apps and deploy workflows

List store categories

GET /store/categories

Returns every marketplace category with display metadata and a live count of public store listings in that category. Categories are shared across app, skill, MCP, and other store surfaces. The web app Apps loads GET /store/categories?type=app for category chips; List approved apps category filters and MCP server tools such as app_list and mcp_list accept the same slugs.

No API key is required. Results are sorted by rank (ascending), then slug (ascending). Categories with no matching public listings still appear with count: 0 so browse UIs can show the full catalog.

Query parameters

Response (StoreCategoryDTO[])

1curl "https://api.inference.sh/store/categories?type=app"

1{2  "data": [3    {4      "slug": "image",5      "name": "Image",6      "description": "Generate, edit and understand images.",7      "icon": "image",8      "rank": 10,9      "count": 12810    },11    {12      "slug": "decision",13      "name": "Decision",14      "description": "Classify, route and score with calibrated probabilities.",15      "icon": "split",16      "rank": 65,17      "count": 418    }19  ]20}

Use this endpoint for category sidebars and filters. Listing category / subcategory fields on GET /store/apps still describe each app row; this endpoint is the canonical category catalog. App objects use a subset of these slugs for category (Apps API — App category). Skill and connector listings may use additional slugs (for example developer, productivity) that appear here but not on every app. Store listings can carry one to three categories when published; app Grid cards show a primary category field derived from the listing. See Browsing the Grid — Categories.

List browseable tags

GET /store/tags

Returns app tag slugs that at least three public apps carry, with a display title and count. The web app Apps loads this list for the tasks tag filter in the explore sidebar (wide screens) and the horizontal tag row on small screens — labels are the slug with hyphens replaced by spaces; the title field is for API clients and other UIs. Tags on fewer public apps are still valid on apps and in tag filters on List approved apps; they are omitted here until usage crosses that bar.

No API key is required. Results are sorted by count (descending), then slug (ascending).

Response (StoreTagDTO[])

1curl "https://api.inference.sh/store/tags"

1{2  "data": [3    { "slug": "text-to-image", "title": "Text to Image", "count": 42 },4    { "slug": "open-weights", "title": "Open Weights", "count": 18 }5  ]6}

Use this endpoint (not generated SDK enums) for tag pickers, browse surfaces, and display titles. Older SDK releases exported AppTag / AppTagTitle constants; those were removed in favor of this API.

List approved apps

GET /store/apps or POST /store/apps/list

Returns apps with an approved store version, sorted by rank (descending) when you omit sort. Apps with status retired are excluded. Other statuses (active, maintenance, deprecated) may appear when the underlying app is public — see Apps API — App lifecycle status.

Query parameters (GET)

POST accepts the same fields in the JSON body.

Filters

Pass filters as a JSON array on GET (filters=...) or in the POST body. Each entry is { "field", "value", "operator" } with operator typically eq.

The web app Apps explore page uses category and tag filters together with text search.

1curl -X POST https://api.inference.sh/store/apps/list \2  -H "Content-Type: application/json" \3  -d '{4    "limit": 20,5    "filters": [6      { "field": "category", "value": "image", "operator": "eq" },7      { "field": "tag", "value": "text-to-image", "operator": "eq" }8    ]9  }'

Response (PublicAppStoreDTO)

1curl "https://api.inference.sh/store/apps?limit=20"

1curl -X POST https://api.inference.sh/store/apps/list \2  -H "Content-Type: application/json" \3  -d '{"limit": 20, "search": {"term": "whisper", "fields": ["name", "description"]}}'

Pagination matches other cursor list endpoints (items, next_cursor, has_next, …).

Get app store listing

GET /store/apps/{appId}

Returns store metadata for one approved app (AppStoreListingDTO). Use this when you already have an app ID from a list response.

1curl "https://api.inference.sh/store/apps/app_abc123"

Reserved cloud capacity (min_concurrency)

When min_concurrency is greater than zero, a worker holding the app is not evicted for another app's task while min_concurrency or fewer workers hold it. It is displacement protection only: the scheduler does not start or pre-load workers to reach the number. An app with no worker holding it gets no protected workers until a task loads it.

It is part of the app's runtime policy, set by the platform per app. Workspaces cannot configure it. See Workers concept and Task failure messages.

Get app pricing

GET /store/apps/{appId}/pricing

Returns the published pricing config for the app's current store version (AppPricing — CEL expressions, price variables, rendered description). Used by public app pages.

1curl "https://api.inference.sh/store/apps/app_abc123/pricing"

→ Usage-based pricing · Get task cost

Estimate app cost

POST /store/apps/{appId}/estimate

Preview run cost before dispatch. Evaluates pricing expressions that only depend on task inputs (task_inputs, prices, fees, task_function). When fee expressions need post-execution data (outputs, elapsed_ms, elapsed_seconds, resource_fee, and so on), the API falls back to the app's estimate CEL expression if one is configured in AppPricing.

No authentication required.

Request body

Response (EstimateCostResponse)

Example (exact estimate):

1curl -X POST "https://api.inference.sh/store/apps/app_abc123/estimate" \2  -H "Content-Type: application/json" \3  -d '{"input": {"n": 4}}'

1{2  "confidence": "exact",3  "microcents": 2000000,4  "pricing_description": "$0.02 per image"5}

Example (range estimate via estimate expression):

1{2  "confidence": "range",3  "min": 150000,4  "max": 1500000,5  "pricing_description": "About $0.0015–$0.015 per run"6}

Example (unknown — output-dependent pricing, no estimate expression):

1{2  "confidence": "unknown",3  "depends_on": ["outputs"]4}

Example (estimate_error — estimate expression failed):

1{2  "confidence": "unknown",3  "estimate_error": "no such key: max_tokens",4  "depends_on": ["outputs"]5}

→ Usage-based pricing — authoring estimate expressions
→ Evaluate CEL pricing — preview draft formulas with sample output_meta

List approved skills

GET /store/skills or POST /store/skills/list

Returns skills with an approved store listing. Same cursor, filter, sort, and search parameters as List approved apps.

Response (PublicSkillStoreDTO)

1curl "https://api.inference.sh/store/skills?limit=20&filters=%5B%7B%22field%22%3A%22namespace%22%2C%22value%22%3A%22inferencesh%22%2C%22operator%22%3A%22eq%22%7D%5D"

Example filters value (URL-encoded above):

1[{"field": "namespace", "value": "inferencesh", "operator": "eq"}]

Get skill store listing

GET /store/skills/{skillId}

Returns store metadata for one skill (SkillStoreListingDTO), including installs and uses counts.

1curl "https://api.inference.sh/store/skills/skill_abc123"

To fetch skill content (instructions and files), use Skills API (GET /skills/{namespace}/{name}/content).

List approved agents

GET /store/agents or POST /store/agents/list

Returns public agents with a live store listing, in listing order, as AgentDTO objects. Same cursor, filter, sort, and search parameters as List approved apps; search reads the agent's name and title. POST /agents/list lists your workspace's agents instead.

1curl "https://api.inference.sh/store/agents?limit=20"

List approved flows

GET /store/flows or POST /store/flows/list

Returns public flows with a live store listing, in listing order, as FlowDTO objects. Same parameters as List approved apps; search reads the flow's name, title, and description. POST /flows/list lists your workspace's flows instead.

1curl "https://api.inference.sh/store/flows?limit=20"

Count listings per type

GET /store/counts

Returns how many public listings each resource type has (app, agent, flow, skill, knowledge, mcp). A type with none is absent. Explore hides the tabs that would list nothing.

1curl "https://api.inference.sh/store/counts"

1{ "app": 446, "skill": 128, "mcp": 37, "agent": 12 }

Listings when a resource is deleted

When you delete an app, skill (knowledge entry), agent, flow, or MCP server, the platform removes its store listing in the same operation. The listing is marked delisted, disappears from publisher admin views and public store browse APIs, and the listing history records that the resource was deleted. Search indexing removes the resource via the normal delete path.

Marketplace listing pages (public)

GET /marketplace/{type}/{namespace}/{name} returns the public detail (MarketplaceDetailDTO) for one live listing. No API key is required. type is app, skill, mcp, agent, flow, or knowledge. Listing slugs are namespace/name for most types; MCP listings use namespace/server_slug.

1curl "https://api.inference.sh/marketplace/app/inferencesh/whisper"

Signed-in callers can report a listing with POST /marketplace/{type}/{namespace}/{name}/report (apps:read).

Publisher marketplace (authenticated)

Most routes need apps:read or apps:write. Publisher profile routes need teams:read / teams:write and the workspace edit profile capability. Every route that changes a listing (draft, edit, submit, review messages, delist, relist, publish, withdraw) also needs edit profile on the listing's workspace, since a listing is the team's public profile; write access on the resource alone returns 403 team_role_required. POST /store/submit requires a person (a signed-in user or their personal key), not a workspace API key.

Publishing invitation

New marketplace publish actions are invitation-only. Before POST /store/listings/draft, POST /store/check, or POST /store/submit, the API checks the workspace entitlement feature:marketplace_publish. A missing or disabled grant returns 403 feature_not_available:

the marketplace takes submissions by invitation for now; join the waitlist to be let in

Reading listings your team already owns (for example GET /store/listings/{id}) is not blocked. Ask for access with Entitlement requests; the web app's publisher waitlist submits the infsh/marketplace_waitlist form for you.

Publish flow

1. GET /store/terms: policy version, acknowledgement checklist, and rule IDs cited by automated findings.
2. GET /store/publisher / PUT /store/publisher: publisher profile (support, privacy, contact email verification via /store/publisher/verify-email/*).
3. POST /store/listings/draft: create or fetch the draft listing for a resource (resource_type, resource_id).
4. PUT /store/listings/{id} (or PATCH): edit listing-owned fields (tagline, categories, tags, URLs, data-handling answers, use cases).
5. POST /store/check: run validation, scan, and policy checks on a resource version without submitting.
6. POST /store/submit: submit a version with changelog, license, and attestation (terms_version, policy_version, data-handling answers, ticked acks from /store/terms).

POST /store/check and GET /store/versions/{id} return a StoreCheckResultDTO: one row per check (validation, scan, policy, …) with status, summary, and findings[]. Each finding has code, severity (block, hold, warning, info), title / message / fix, an optional policy rule, and an optional key naming the resource field to change. A version migrated from the old store tables (origin: migrated) reports each check it never ran with status skipped ("not run: migrated version").

Store vs search vs resource APIs

Related