- actions_required
- entity:view
- entity:update
- file:view
- file:update
- description
- Generates rich descriptions for entities based on their metadata and relationships
- endpoint
- https://description-service.arke.institute
- endpoint_verified_at
- 2026-01-21T02:15:45.822Z
- input_schema
- properties
- entity_id
- description
- Entity to describe
- type
- string
- options
- description
- Agent-specific options
- properties
- custom_prompt
- description
- Additional instructions for the LLM
- type
- string
- max_related
- description
- Max related entities to fetch for context (default: 100)
- type
- number
- ordered_count
- description
- First N relations fetched in-order, rest randomly sampled (default: 20)
- type
- number
- predicates
- description
- Which relationship predicates to follow (default: all)
- type
- array
- type
- object
- type
- object
- output_description
- Modifies the target entity in-place by adding four properties. 'description' contains a markdown-formatted encyclopedia-style description (typically 150-300 words) structured with a title heading, an Overview section (what the entity is), a Context section (provenance and background drawn from related entities), and a Contents section (subjects and details). The description may reference related entities using arke: URI links like [Label](arke:entity-id). 'description_title' contains just the title text extracted from the first heading. 'description_generated_at' is an ISO 8601 timestamp of when the description was generated. 'description_model' records which LLM model produced the description. The service gathers context by fetching the target entity's full manifest and up to 100 related entities (first 20 in document order, remainder randomly sampled), feeding all of this as raw JSON to the LLM. No new entities or files are created.
- output_relationships
- This service does not create, modify, or delete any relationships.
- It reads the target entity's existing relationships to discover related entities, which are fetched and provided as context to the LLM.
- By default all relationship predicates are followed; the 'predicates' option can restrict which predicates are used for context gathering.
- The generated description text may contain arke: URI references to related entities (e.g. [Label](arke:entity-id)), but these are inline markdown links within the description property, not structural relationships.
- output_tree_example
- Before:
entity 'Annual Report 2023'
properties: { label: 'Annual Report 2023', filename: 'report.pdf', ... }
After:
entity 'Annual Report 2023'
properties: {
label: 'Annual Report 2023',
filename: 'report.pdf',
...,
description: '# Annual Report 2023\n## Overview\nThis is a PDF document...\n## Context\nPart of [Corporate Records](arke:01ABC123)...\n## Contents\nCovers fiscal year...',
description_title: 'Annual Report 2023',
description_generated_at: '2025-01-15T10:30:00.000Z',
description_model: 'Qwen/Qwen3-235B-A22B-Instruct-2507'
}
- status
- active