> ## Documentation Index
> Fetch the complete documentation index at: https://api.aodocs.com/llms.txt
> Use this file to discover all available pages before exploring further.
# Train Extraction
> Train the metadata extraction rules
## OpenAPI
````yaml https://document-assistant.us.aodocs.app/openapi.json post /assistant/api/v1/tenants/{tenant}/metadata:train
openapi: 3.1.0
info:
title: Document Assistant API
description: |
The Document Assistant API enables developers to integrate document-based assistant
functionalities into their applications using AODocs.
The first step is to enable AIDA (AODocs Intelligent Document Assistant) in your tenant.
The support team will provide you with the necessary information on how to get started.
The second step is to enable AIDA on libraries. You need to be an administrator of the
library to do this.
AIDA can answer questions only from those documents that are in an AIDA enabled
library.
We create a vector store index for these documents. This can take some time. The get
library API provides the status of the indexation and an estimate of the time
remaining.
The content of the index is automatically kept up-to-date near real time with the
content changes.
The third step is to ask questions based on the AODocs documents.
AIDA automatically filters the results based on the user's permissions.
Authentication: All API endpoints require a valid OAuth2 token in the
Authorization header. Ensure the client ID of the OAuth2 token is whitelisted in AODocs.
version: 1.1.0
servers:
- url: https://document-assistant.us.aodocs.app
description: Production US
- url: https://document-assistant.eu.aodocs.app
description: Production EU
security: []
paths:
/assistant/api/v1/tenants/{tenant}/metadata:train:
post:
tags:
- metadata
summary: Train Extraction
description: Train the metadata extraction rules
operationId: train_extraction_assistant_api_v1_tenants__tenant__metadata_train_post
parameters:
- name: tenant
in: path
required: true
schema:
anyOf:
- type: string
- type: 'null'
description: The domain of the customer
examples:
- example.com
title: Tenant
description: The domain of the customer
requestBody:
required: true
content:
application/json:
schema:
$ref: '#/components/schemas/MetadataExtractRequest'
responses:
'200':
description: Successful Response
content:
application/json:
schema:
$ref: '#/components/schemas/ExtractionSuggestion'
'422':
description: Validation Error
content:
application/json:
schema:
$ref: '#/components/schemas/HTTPValidationError'
security:
- HTTPBearer: []
components:
schemas:
MetadataExtractRequest:
properties:
llm_provider:
anyOf:
- $ref: '#/components/schemas/LlmProvider'
- type: 'null'
description: >-
The LLM provider to use. If not provided, the default provider will
be used.
llm_model_prefix:
anyOf:
- type: string
- type: 'null'
title: Llm Model Prefix
description: >-
The prefix of the LLM model name to use. If not provided, the
default model will be used.
examples:
- gemini-2.5-pro
- gemini-2.5-pro-preview-05-06
- gpt-4.1
document_id:
type: string
title: Document Id
description: The document ID of the files
attachment_filters:
anyOf:
- items:
$ref: '#/components/schemas/AttachmentFilter'
type: array
- type: 'null'
title: Attachment Filters
description: |-
The attachments to extract metadata from.
If provided then only the attachments matching one of the rules are used.
If not provided then all attachments will be extracted.
Deprecated: use sources with an attachments entry instead.
deprecated: true
sources:
anyOf:
- items:
oneOf:
- $ref: '#/components/schemas/DescriptionExtractionSource'
- $ref: '#/components/schemas/AttachmentsExtractionSource'
discriminator:
propertyName: type
mapping:
attachments:
$ref: '#/components/schemas/AttachmentsExtractionSource'
description:
$ref: '#/components/schemas/DescriptionExtractionSource'
type: array
minItems: 1
- type: 'null'
title: Sources
description: |-
The ordered document content sources to extract metadata from.
If omitted, attachment-only extraction is used for backward compatibility.
Description sources are opt-in.
instruction:
anyOf:
- type: string
- type: 'null'
title: Instruction
description: The prompt extension for the metadata extraction
provide_location:
type: boolean
title: Provide Location
description: >-
Whether to calculate and return page locations (bounding boxes) for
the quote supporting each extracted field. Quotes are matched
against textual and scanned PDF attachments plus PNG, JPEG, and WebP
images without changing extracted values. Unsupported attachment
types and quotes that do not occur verbatim after normalization do
not produce boxes.
default: false
fields:
anyOf:
- items:
$ref: '#/components/schemas/FieldSpec'
type: array
- items:
type: string
type: array
- type: 'null'
title: Fields
description: >-
Specification which metadata to extract and how. If not provided,
all fields will be extracted.
example_documents:
anyOf:
- items:
type: string
type: array
- type: 'null'
title: Example Documents
description: >-
A list of example documents to use for metadata extraction. If
provided, the LLM will use these documents to better understand the
context and provide more accurate extractions.
registries:
anyOf:
- items:
$ref: '#/components/schemas/Registry'
type: array
- type: 'null'
title: Registries
description: >-
A list of metadata registries to use for metadata extraction. If
provided, the LLM will use these registries to better understand the
context and provide more accurate extractions.
type: object
required:
- document_id
title: MetadataExtractRequest
ExtractionSuggestion:
properties:
fields:
items:
$ref: '#/components/schemas/FieldSuggestion'
type: array
title: Fields
description: The suggested instructions for extracting the fields
extraction_instruction:
type: string
title: Extraction Instruction
description: The suggested instruction for the whole document
comment:
anyOf:
- type: string
- type: 'null'
title: Comment
description: Reason for the instruction change
type: object
required:
- fields
- extraction_instruction
- comment
title: ExtractionSuggestion
HTTPValidationError:
properties:
detail:
items:
$ref: '#/components/schemas/ValidationError'
type: array
title: Detail
type: object
title: HTTPValidationError
LlmProvider:
type: string
enum:
- OpenAI
- Google
- Altirnao
- MistralAI
- Anthropic
title: LlmProvider
AttachmentFilter:
properties:
name:
anyOf:
- type: string
- type: 'null'
title: Name
description: The name of the attachment to match (it can be a regular expression)
id:
anyOf:
- type: string
- type: 'null'
title: Id
description: The id of the matched attachment
mime_type:
anyOf:
- type: string
- type: 'null'
title: Mime Type
description: The mimetype of the matched attachment
type: object
title: AttachmentFilter
description: Match an attachment by identifier, MIME type, or name expression.
DescriptionExtractionSource:
properties:
type:
type: string
const: description
title: Type
default: description
type: object
title: DescriptionExtractionSource
AttachmentsExtractionSource:
properties:
type:
type: string
const: attachments
title: Type
default: attachments
include:
anyOf:
- items:
$ref: '#/components/schemas/AttachmentFilter'
type: array
- type: 'null'
title: Include
description: >-
Attachment filters to include. If omitted, all attachments are
included. If empty, no attachments are included.
exclude:
anyOf:
- items:
$ref: '#/components/schemas/AttachmentFilter'
type: array
- type: 'null'
title: Exclude
description: >-
Attachment filters to exclude after applying include filters.
Exclusion takes precedence over inclusion.
type: object
title: AttachmentsExtractionSource
FieldSpec:
properties:
name:
anyOf:
- type: string
- type: 'null'
title: Name
description: The name of the metadata field. Either the name or id is mandatory
id:
anyOf:
- type: string
- type: 'null'
title: Id
description: The id of the metadata field. Either the name or id is mandatory
instruction:
anyOf:
- type: string
maxLength: 1024
- type: 'null'
title: Instruction
description: Custom instruction for extracting this field
barcode:
type: boolean
title: Barcode
description: >-
Extract decoded barcode content and locations without using the LLM.
Supported only for STRING and TEXT fields.
default: false
integer_decimal_handling:
$ref: '#/components/schemas/IntegerDecimalHandling'
description: >-
How to coerce a decimal value the LLM returns for this integer-typed
field: `truncate` drops the fractional part (3.7 → 3), `round`
rounds to the nearest integer (3.5 → 4, 3.4 → 3), `roundup` always
rounds up toward positive infinity (3.2 → 4), `error` raises an
extraction error (default — current behavior). Applies only to
integer fields; ignored otherwise.
default: error
table_search:
anyOf:
- $ref: '#/components/schemas/TableSearchConfig'
- type: 'null'
description: >-
Optional request-scoped table-search configuration used to validate
and canonicalize STRING, CATEGORY, REFERENCE_CATALOG, or TABLE
values. Complete low-cardinality enum values remain available to the
extraction model; incomplete or high-cardinality enums are extracted
as plain text.
type: object
title: FieldSpec
Registry:
properties:
name:
type: string
title: Name
description: >-
The name of the metadata registry. This is used to identify the
registry in the request.
library_id:
type: string
title: Library Id
description: >-
The library ID that contains the metadata registry that should be
used for metadata extraction.
document_class_id:
type: string
title: Document Class Id
description: >-
The document class ID that contains the metadata registry that
should be used for metadata extraction.
match_mode:
$ref: '#/components/schemas/MatchMode'
description: >-
Specifies whether all key fields must match (ALL) or if any single
key field match is sufficient (ANY) to consider a registry entry as
a potential match.
default: ALL
field_mappings:
items:
$ref: '#/components/schemas/FieldMapping'
type: array
title: Field Mappings
description: |-
Mapping between the extracted fields and the registry
There should be at least one key field in this mapping.
type: object
required:
- name
- library_id
- document_class_id
- field_mappings
title: Registry
FieldSuggestion:
properties:
name:
type: string
title: Name
description: The name of the field
extraction_instruction:
type: string
title: Extraction Instruction
description: The suggested instruction for extracting this field
comment:
anyOf:
- type: string
- type: 'null'
title: Comment
description: >-
Reason for the instruction change. You can argue here if the
expected value is wrong and no reasonable instruction can be
provided.
type: object
required:
- name
- extraction_instruction
- comment
title: FieldSuggestion
ValidationError:
properties:
loc:
items:
anyOf:
- type: string
- type: integer
type: array
title: Location
msg:
type: string
title: Message
type:
type: string
title: Error Type
input:
title: Input
ctx:
type: object
title: Context
type: object
required:
- loc
- msg
- type
title: ValidationError
IntegerDecimalHandling:
type: string
enum:
- truncate
- round
- roundup
- error
title: IntegerDecimalHandling
TableSearchConfig:
properties:
dataset_id:
type: string
minLength: 1
title: Dataset Id
description: Tenant-scoped table-search dataset ID.
input_columns:
anyOf:
- items:
type: string
type: array
minItems: 1
- type: 'null'
title: Input Columns
description: >-
Dataset filter columns. Wrap a single STRING filter in a one-item
list.
output_columns:
anyOf:
- items:
$ref: '#/components/schemas/TableSearchOutputColumn'
type: array
minItems: 1
- type: 'null'
title: Output Columns
description: >-
Dataset output columns. Wrap a single STRING output in a one-item
list.
type: object
required:
- dataset_id
title: TableSearchConfig
MatchMode:
type: string
enum:
- ALL
- ANY
title: MatchMode
FieldMapping:
properties:
extracted_field:
type: string
title: Extracted Field
description: The id of the field in the extracted metadata
registry_field:
type: string
title: Registry Field
description: The id of the field in the metadata registry
key_field:
type: boolean
title: Key Field
description: >-
If true, the field is used to match the document to the metadata
registry.
We need at least one key field to match the document to the metadata
registry.
default: false
optional_key:
type: boolean
title: Optional Key
description: >-
If True, this key field is optional—if it's missing in the extracted
data, the match can still succeed as long as other key fields are
present and match.
default: false
similarity_threshold:
type: number
maximum: 1
minimum: 0
title: Similarity Threshold
description: >-
For key fields only: how closely the extracted and registry values
need to match (0.0 = no similarity, 1.0 = perfect match). If more
than one record matches, the one with the highest total similarity
across all key fields will be chosen.
default: 1
type: object
required:
- extracted_field
- registry_field
title: FieldMapping
TableSearchOutputColumn:
properties:
name:
type: string
minLength: 1
title: Name
description: >-
Dataset output column. For TABLE fields this must exactly match the
metadata table-column name.
id_column:
anyOf:
- type: string
minLength: 1
- type: 'null'
title: Id Column
description: >-
Companion dataset output column containing the category or reference
catalog value ID. It is not exposed as a separate extracted column.
type: object
required:
- name
title: TableSearchOutputColumn
securitySchemes:
HTTPBearer:
type: http
scheme: bearer
````