Search

Search for documents matching a specific query in the given index.

POST /indexes/{index_name}/search

Path parameters

Name	Type	Description
`index_name`	String	name of the requested index

Body

The body parameters below would be used for HTTP requests (if you were using cURL, for example). Python client users should use the pythonic snakecase equivalents (for example, searchable_attributes rather than searchableAttributes).

Search Parameter	Type	Default value	Description
`q`	String OR Dict	`null`	Query string, weighted query strings (if Dict). Optional for tensor search if context parameter is used.
`limit`	Integer	`10`	Maximum number of documents to be returned
`offset`	Integer	`0`	Number of documents to skip (used for pagination)
`filter`	String	`null`	Filter string in the Marqo DSL Language. In the Python client this parameter is called `filter_string`: `mq.search("my query", filter_string="country:(United States)")`
`searchableAttributes`	Array of strings	`null`	Attributes to be queried during the search. Only supported in structured indexes
`showHighlights`	Boolean	`true`	Return highlights for the document match. Only applicable for TENSOR search. With LEXICAL search, highlights will always be `[]`.
`searchMethod`	String	`"TENSOR"`	The search method, can be LEXICAL or TENSOR
`attributesToRetrieve`	Array of strings	`null`	Attributes to return in the search response
`efSearch`	Integer	`2000`	efSearch is the size of the dynamic list for the nearest neighbors (used during the search) - higher gives better recall at the cost of latency. Also efSearch must be greater than limit and limit is capped at 400
`approximate`	Boolean	`True`	Approximate toggles between exact KNN and approximate KNN (with HNSW)
`reRanker`	String	`null`	Method to use for reranking results
`imageDownloadHeaders`	Dict	`{}`	Headers for the image download. Can be used to authenticate the images for download.
`context`	Dict	`null`	Dictionary of `"tensor":{List[{"vector": List[floats], "weight": (float)}]}` to bring your own vectors into search.
`scoreModifiers`	Dict	`null`	A dictionary to modify the score based on field values. Check here for examples.
`modelAuth`	Dict	`null`	Authorisation details used by Marqo to download non-publicly available models. Check here for examples.
`textQueryPrefix`	String	`null`	The prefix added to text queries when embedding. This field overrides the `textQueryPrefix` set in the index settings during index creation. If it unset by the user, it defaults to the prefixes defined in the index settings. For more information on default values for index settings, see create_index.

Note on Attributes to Retrieve per Query

It is beneficial to explicitly set the attributesToRetrieve parameter to limit the amount of data Marqo returns per document. Latency will increase as the number of attributes and documents retrieved increases. If you have documents with many fields that are not used by systems interfacing with Marqo's results, setting attributesToRetrieve to the minimal set of fields required can reduce latency and improve throughput.

Query parameters

Search Parameter	Type	Default value	Description
`device`	String	`null`	The device used to search. If `device` is not specified and CUDA devices are available to Marqo (see here for more info), Marqo will speed up search by using an available CUDA device. Otherwise, the CPU will be used. Options include `cpu` and `cuda`, `cuda1`, `cuda2` etc. The `cuda` option tells Marqo to use any available cuda devices.
`telemetry`	Boolean	`False`	If true, the `telemetry` object is returned in the search response body. This includes information like latency metrics. This is set at client instantiation time in the Python client: `mq = marqo.Client(return_telemetry=True)`

Search result pagination

Use parameters limit and offset to paginate your results, meaning to query a certain number of results at a time instead of all at once.

The limit parameter sets the size of a page. If you set limit to 10, Marqo's response will contain a maximum of 10 search results. The offset parameter skips a number of search results. If you set offset to 20, Marqo's response will skip the first 20 search results.

Let's say you want each page to have 10 results, and you want to receive the 2nd page. Try setting limit and offset like so:

# Specify page properties
page_size = 10
page_num = 2

# Set limit and offset accordingly
limit = page_size
offset = (page_num - 1) * page_size

Pagination limitations

Search results can only be 10,000 results deep. This means limit + offset must be less than or equal to 10000. Also, efSearch must be greater than limit+offset.

Using pagination with search_method="TENSOR" may result in some results being skipped or duplicated (often near the edge of pages) within the first few pages if the page size is much smaller than the total search result count. Please keep this in mind when looking for particular results or when result order is essential.

Lexical search: exact matches

Use searchMethod="LEXICAL" to perform keyword search instead of tensor search. With lexical search, you can enable exact match searching using double quotes: "".

Any term enclosed in "" will be labeled a required term, which must exist in at least one field of every result hit. Note that terms enclosed in double quotes must also have a space between them and the terms before and after them, same as regular terms. Use this feature to filter your results to only documents containing certain terms. For example, if you want to search for results containing fruits, vegetables, or candy, but they must be green, you can construct your query as such:

Python

mq.index("my-first-index").search(
    q='fruit vegetable candy "green"',
    search_method="LEXICAL"
)

If you want to escape the double quotes (interpret them as text), use 2 escape keys \\. For example: q = 'Dwayne \\"The Rock\\" Johnson'.

Note: syntax errors

If your use of "" does not follow proper syntax, the entire query will simply be interpreted literally, with no required terms. Here some examples of syntax errors:

# Quoted terms without spaces before/after
q = 'apples"oranges" bananas'
q = 'cucumbers "melons and watermelons""grapefruit"'

# Unescaped quotes
q = 'There is a quote right"here'

# Unbalanced quotes
q = '"Dr. Seuss" "Thing 1" "Thing 2'

Response

Name	Type	Description
`hits`	Array of objects	Results of the query
`limit`	Integer	Number of documents chunks specified in the query
`offset`	Integer	Number of skipped results specified in the query
`processingTimeMs`	Number	Processing time of the query
`query`	String	Query originating the response

Example

cURLPython

curl -XPOST 'http://localhost:8882/indexes/my-first-index/search' -H 'Content-type:application/json' -d '
{
    "q": "what is the best outfit to wear on the moon?",
    "limit": 10,
    "offset": 0,
    "showHighlights": true,
    "searchMethod": "TENSOR",
    "attributesToRetrieve": ["Title", "Description"]
}'

mq.index("my-first-index").search(
    q="What is the best outfit to wear on the moon?",
    limit=10,
    offset=0,
    show_highlights=True,
    search_method="LEXICAL",
    attributes_to_retrieve=["Title", "Description"]
)

Response: `200 Ok`

{
  "hits": [
    {
      "Title": "Extravehicular Mobility Unit (EMU)",
      "Description": "The EMU is a spacesuit that provides environmental protection, mobility, life support, and communications for astronauts",
      "_highlights": [
        {
          "Description": "The EMU is a spacesuit that provides environmental protection, mobility, life support, and communications for astronauts"
        }
      ],
      "_id": "article_591",
      "_score": 1.2387788
    },
    {
      "Title": "The Travels of Marco Polo",
      "Description": "A 13th-century travelogue describing Polo's travels",
      "_highlights": [
        {
          "Title": "The Travels of Marco Polo"
        }
      ],
      "_id": "e00d1a8d-894c-41a1-8e3b-d8b2a8fce12a",
      "_score": 1.2047464
    }
  ],
  "limit": 10,
  "offset": 0,
  "processingTimeMs": 49,
  "query": "What is the best outfit to wear on the moon?"
}

Query (q)

Parameter: q

Expected value: Search string, a dictionary of weighted search strings. Optional for tensor search if context parameter is used.

Search strings are text or a url to an image, if the index has treatUrlsAndPointersAsImages set to True.

If queries are weighted, each weight act as a (possibly negative) multiplier for that query, relative to the other queries.

If your search method is TENSOR, this parameter is optional if you are using the context parameter. At least one of q or context must be specified for this search.

Default value: null

Examples

# query string: 
q = "How do I keep my plant alive?"

# a dictionary of weighted query strings
q = {
    # a weighting of 1 gives this query a neutral effect:
    "Which dogs are the best pets": 1.0,
    # we give this a weighting of 2 because we really want results similar to this:
    "https://image_of_a_golden_retriever.png": 2.0,
    # we give this a negative weighting to make it less likely to appear: 
    "Poodle": -1
}

Limit

Parameter: limit

Expected value: Any positive integer

Default value: 10

Max: 1000

Sets the maximum number of documents returned by a single query.

Offset

Parameter: offset

Expected value: Any integer greater than or equal to 0

Default value: 0

Max: 10000

Sets the number of documents to skip. For example, if offset = 20, The first result returned will be the 21st result. Only set this parameter for single-field searches (multi-field support to follow).

Filter

Parameter: filter

Expected value: A filter string written in Marqo's query DSL.

Default value: null

Uses filter expressions to refine search results.

Read our guide on filtering, faceted search and filter expressions.

Example

You can write a filter expression in string syntax using logical connectives (see filtering in Marqo):

"(type:confectionary AND food:(ice cream)) OR animal:hippo"

Searchable attributes

Parameter: searchableAttributes

Expected value: An array strings

Default value: null

Configures which attributes will be searched for query matches. This field is only supported in structured indexes.

If no value is specified, all fields will be searched.

Example

You can write the searchableAttributes as a list of strings, for example if you only wanted to search the "Description" field of your documents:

["Description"]

Reranker

Parameter: reRanker

Expected value: One of "owl/ViT-B/32", "owl/ViT-B/16", "owl/ViT-L/14"

Default value: null

Selects the method for reranking results. See the Models reference reranking section for more details.

If no value is specified, reRanker will be set to null and no reranking will occur.

Example

You can write reRanker as a string, for example:

"owl/ViT-B/32"

Context

Parameter: context

Expected value: Dictionary of "tensor":{List[{"vector": List[floats], "weight": (float)}]}

Default value: null

Context allows you to use your own vectors as context for your queries. Your vectors will be incorporated into the query using a weighted sum approach, allowing you to reduce the number of inference requests for duplicated content. The dimension of the provided vectors should be consistent with the index dimension.

Example

mq.index("my-first-index").search(
    q={"Chocolate chip cookies": 1},
    # the dimension of the vector (which is 768 here) should match the dimension of the index
    context={"tensor": [{"vector": [0.3, ] * 768, "weight": 2},  # custom vector 1
                        {"vector": [0.12, ] * 768, "weight": -1}, ]  # custom vector 2
             }
)

Score modifiers

Parameter: score_modifiers

Expected value: An object with two optional keys: multiply_score_by and add_to_score. The value of each of these keys is an array of objects that each contain the name of a numeric field in the document as the field_name key and the weighting that should be applied to the numeric value, as the weight key, if it is found in the doc. If the score modifier field in the document is a map, access the subfield value using dot notation.

Default value: null

Score modifiers allows you to modify the initial score of the document by multiplying, and adding to, the initial search with values found within the document itself. This allows you to modify the search results based on metadata not included in the vectors.

The default weight value is 1 in the multiply_score_by object and 0 in the add_to_score object. The multiply_score_by modifiers will be applied to the document's score before the add_to_score modifiers. If a field specified in the score modification objects isn't found in the document, then the score modification will be skipped for that document's field.

For map score modifiers, avoid retrieving the score modifier fields in the query if they are not necessary for retrieval. For more information, see attributesToRetrieve.

There is negligible performance impact in performing queries with 1000 score modifiers against large dictionaries of upwards of 15,000 score modifiers per document.

Example

mq.index("my-first-index").add_documents(
    documents=[
        {
            "productImage": "https://my-images.com/cool-tshirt-1.png",
            "itemPopularity": 2.1,
            "negativeReviewCount": 4
        }],
    tensor_fields=['productImage']
)

mq.index("my-first-index").search(
    q="T-shirts with a cartoon character",
    score_modifiers={
        "multiply_score_by": [{"field_name": "itemPopularity", "weight": 1.8}],
        "add_to_score": [{"field_name": "negativeReviewCount", "weight": -0.1}]
    }
)

# if the initial score of the search query against this document is 0.67, then, after applying score modifiers,
# it will be modifed to 0.67 * (1.8 * 2.1) + (-0.1 * 4) = 2.13

Example Using Map Score Modifiers

docs = [
    {"_id": "1", "text_field": "a photo of a cat", "map_score_mods": {"a": 0.5}},
    {"_id": "2", "text_field": "a photo of a dog", "map_score_mods": {"b": 0.5}},
    {"_id": "3", "text_field": "a photo of a cat", "map_score_mods": {"c": 0.5}},
    {"_id": "4", "text_field": "a photo of a cat", "map_score_mods_int": {"a": 1}},
    {"_id": "5", "text_field": "a photo of a cat", "map_score_mods_int": {"b": 1}},
    {"_id": "6", "text_field": "a photo of a cat", "map_score_mods_int": {"c": 1}},
    {"_id": "7", "text_field": "a photo of a cat", "map_score_mods_int": {"c": 1}, "map_score_mods": {"a": 0.5}},
    {"_id": "8", "text_field": "a photo of a dog", "my_int": 2},
]

res = mq.index("my-unstructured-index").add_documents(
    documents=docs, 
    tensor_fields=["text_field"],
)

# The same search sytax is used for both structured and unstructured indexes
res = mq.index("map-score-modifiers-index").search(
    q="",
    score_modifiers={
        "add_to_score": [{"field_name": "map_score_mods_int.c", "weight": 2}],
        "multiply_score_by": [{"field_name": "map_score_mods.a", "weight": 4}]
    },
    attributes_to_retrieve=["_id", "text_field"]
)
print(json.dumps(res, indent=2))

Model Auth

Parameter: modelAuth

Expected value: Dictionary with either an s3 or an hf model store authorisation object.

Default value: null

The ModelAuth object allows searching on indexes that use OpenCLIP and CLIP models from private Hugging Face and AWS S3 stores.

The modelAuth object contains either an s3 or an hf model store authorisation object. The model store authorisation object contains credentials needed to access the index's non publicly accessible model. See the example for details.

The index's settings must specify the non publicly accessible model's location in the setting's modelProperties object.

ModelAuth is used to initially download the model. After downloading, Marqo caches the model so that it doesn't need to be redownloaded.

Example: AWS s3

# Create an index that specifies the non-public location of the model.
# Note the `auth_required` field in `modelProperties` which tells Marqo to use
# the modelAuth it finds during search to download the model
mq.create_index(
    index_name="my-cool-index",
    settings_dict={
        "treatUrlsAndPointersAsImages": True,
        "model": 'my_s3_model',
        "normalizeEmbeddings": True,
        "modelProperties": {
            "name": "ViT-B/32",
            "dimensions": 512,
            "modelLocation": {
                "s3": {
                    "Bucket": "<SOME BUCKET>",
                    "Key": "<KEY TO IDENTIFY MODEL>",
                },
                "authRequired": True
            },
            "type": "open_clip",
        }
    }
)

# Specify the authorisation needed to access the private model during search:
# We recommend setting up the credential's AWS user so that it has minimal 
# accesses needed to retrieve the model
mq.index("my-cool-index").search(
    q="Chocolate chip cookies",
    model_auth={
        's3': {
            "aws_access_key_id": "<SOME ACCESS KEY ID>",
            "aws_secret_access_key": "<SOME SECRET ACCESS KEY>"
        }
    }
)

Example: Hugging Face (HF)

# Create an index that specifies the non-public location of the model.
# Note the `auth_required` field in `modelProperties` which tells Marqo to use
# the modelAuth it finds during search to download the model
mq.create_index(
    index_name="my-cool-index",
    settings_dict={
        "treatUrlsAndPointersAsImages": True,
        "model": 'my_hf_model',
        "normalizeEmbeddings": True,
        "modelProperties": {
            "name": "ViT-B/32",
            "dimensions": 512,
            "modelLocation": {
                "hf": {
                    "repoId": "<SOME HF REPO NAME>",
                    "filename": "<THE FILENAME TO DOWNLOAD>",
                },
                "authRequired": True
            },
            "type": "open_clip",
        }
    }
)

# specify the authorisation needed to access the private model during search:
mq.index("my-cool-index").search(
    q="Chocolate chip cookies",
    model_auth={
        'hf': {
            "token": "<SOME HF TOKEN>",
        }
    }
)

Query Prefixes

Parameters: textQueryPrefix

Expected value: A string.

Default value: ""

This field overrides the text query prefix set during the index's creation.

Note: Users do not need to provide textQueryPrefix for e5 models unless you want to override our default prefixes.

Example: Adding prefixes to search queries. Overriding index defaults

cURLPython

curl -XPOST 'http://localhost:8882/indexes/{index_name}/search' \
-H 'Content-type:application/json' -d '
{
    "q": "Men shoes brown",
    "textQueryPrefix": "override query: "
}'

mq.index("{index_name}").search(
    q="Men shoes brown", text_query_prefix="override query: "
)

Search

Path parameters

Body

Note on Attributes to Retrieve per Query

Query parameters

Search result pagination

Lexical search: exact matches

Response

Example

Response: 200 Ok

Query (q)

Limit

Offset

Filter

Example

Searchable attributes

Example

Reranker

Example

Context

Example

Score modifiers

Example

Example Using Map Score Modifiers

Model Auth

Example: AWS s3

Example: Hugging Face (HF)

Query Prefixes

Example: Adding prefixes to search queries. Overriding index defaults

Subscribe to our mailing list

Response: `200 Ok`