search

Analysis

Search Assets, Models & Workflows

Search for assets (images, video, 3D, audio), models, or workflows by text, visual similarity, or filters. Write-class: inline images in image or images.like/unlike are saved as assets in the selected project before similarity search (deduplicated when possible). Text and existing asset IDs do not upload images. OAuth requires explicit team_id and project_id; no automatic scope selection. Provide query, filter, non-empty filters, image or images. public=true alone does not browse a catalog. Workflows take keyword text and filters only, no image or semantic search. Search hits are summaries — call workflow_get with a hit's id for inputs and the editable graph. Args: - team_id, project_id: required for OAuth callers - query_semantic_ratio: 0.5–0.8 for hybrid keyword + semantic search - filters: structured filters — kind, model_id, type, tags, created_after, created_before, privacy, mime_type, collection_ids - sort_by: ignored when query_semantic_ratio > 0 (semantic search) Returns: matching results with metadata. To view an asset, call asset_display with the asset ID — do NOT try to render asset URLs directly (they are too large for inline display). Examples: - "Find 3D voxel assets" -> target="assets", query="voxel", filters={kind: "3d"} - "Find assets like these but not like that" -> target="assets", images={like: ["<good_id_1>", "<good_id_2>"], unlike: ["<bad_id>"]} - "Search public anime models" -> target="models", query="anime style", public=true - "Recent large 3D assets" -> target="assets", filters={kind: "3d"}, filter='width > 512', sort_by=["createdAt:desc"] - "Find public image apps" -> target="workflows", query="image", public=true, filter='status = "ready"', limit=3 (status uses raw filter, not filters) Don't use when you already know the asset_id or model_id: use asset_get, models_list or asset_display (assets_list and model_get are catalog-only, via scenario_tool_execute_read). To find MCP tools (not platform content), use scenario_tools_search.

open-world

Parameters

Name	Type	Required	Description
target	enum(assets \| models \| workflows)	✓	What to search: assets, models, or workflows. Workflows support keyword text and filters only; provide query, filter, or non-empty filters.
query	string	—	Text search query.
image	string	—	Asset ID, data URL or base64 image for visual similarity. Inline bytes are saved in the selected project. HTTP(S) URLs are unsupported. Mutually exclusive with images.
images	unknown	—	Multi-image similarity with positive (like) and negative (unlike) examples. Mutually exclusive with image.
public	boolean	—	Whether to include public results. true searches public, false or omitted searches private.
filters	unknown	—	Structured filter criteria — converted to Meilisearch filter syntax. Only the listed fields are valid. Use "kind" for asset type (not "modality").
filter	string	—	Raw Meilisearch filter expression (e.g. 'kind = "3d" AND width > 512'). Use only filterable fields for the selected target, with camelCase names such as modelId or mimeType. No regex or id filters. Workflows support status here (e.g. 'status = "ready"'), not in filters. Combined with filters using AND.
query_semantic_ratio	number	—	Balance between keyword (0) and semantic (1) text search. Default 0 (keyword only).
image_semantic_ratio	number	—	Image similarity method: 0 = feature matching, 1 = semantic embedding. Default 1.
sort_by	array	—	Sort attributes, e.g. ["createdAt:desc"].
limit	number	20	Results per page (1-100).
offset	number	0	Pagination offset.
include_urls	boolean	false	true includes signed CDN URLs (url, thumbnail) on each hit (~733 chars each). Default false — asset hits: use asset_download / asset_get when you need a URL; model thumbnails have no URL-free alternative.
team_id	string	—	Team ID. Required with project_id for OAuth callers; an API key is scoped by the key.
project_id	string	—	Project ID. Required with team_id for OAuth callers; an API key is scoped by the key.
response_format	enum(json \| markdown)	json	Output format: 'json' for structured data, 'markdown' for human-readable text.

Example Request

JSON

{
  "target": "assets",
  "query": "dragon concept art",
  "filters": {
    "modality": "image",
    "tags": [
      "dragon",
      "fantasy"
    ]
  },
  "limit": 10,
  "team_id": "team_abc123",
  "project_id": "proj_xyz789"
}

Example Response

JSON

{
  "target": "assets",
  "assets": [
    {
      "id": "asset_img001",
      "name": "dragon_sunset.png",
      "tags": [
        "dragon",
        "fantasy",
        "sunset"
      ],
      "mimeType": "image/png"
    },
    {
      "id": "asset_img002",
      "name": "dragon_cave.png",
      "tags": [
        "dragon",
        "fantasy",
        "cave"
      ],
      "mimeType": "image/png"
    }
  ],
  "estimatedTotalHits": 47
}

Common Use Cases

Find previously generated assets matching a concept to avoid regenerating
Search models by modality (image, video, 3d, audio) to discover available generators
Run visual similarity search by passing a reference image instead of a text query
Filter assets by tag or model to review the output of a specific generation run