> ## Documentation Index
> Fetch the complete documentation index at: https://developers.harvey.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Completion
> Harvey's Completion API provides access to Harvey's models via our orchestration engine, allowing users to ask complex, freeform questions and receive detailed legal-grade responses programmatically. It can also be used in conjunction with files attached to the request, Vaults, and Regional Knowledge Sources.
Access to this API requires `Assistant access via API` permission, which is only available in your workspace with an additional purchase. To get started, contact your Account Manager or reach out to us at support@harvey.ai to discuss an agreement.
### Permissions
Requires `Assistant access via API` and `Assistant access` permission.
## OpenAPI
````yaml /v2_completion_api.json post /api/v2/completion
openapi: 3.0.1
info:
title: v2 - Harvey Assistant Completion API
description: >-
Harvey's Completion API provides access to Harvey's models via our
orchestration engine, allowing users to ask complex, freeform questions and
receive detailed legal-grade responses programmatically. It can also be used
in conjunction with files attached to the request, Vaults, and Regional
Knowledge Sources.
version: 1.0.0
contact:
email: support@harvey.ai
servers:
- url: https://api.harvey.ai
description: Harvey API server
security: []
paths:
/api/v2/completion:
post:
tags:
- Completion
summary: Completion
description: >-
Harvey's Completion API provides access to Harvey's models via our
orchestration engine, allowing users to ask complex, freeform questions
and receive detailed legal-grade responses programmatically. It can also
be used in conjunction with files attached to the request, Vaults, and
Regional Knowledge Sources.
Access to this API requires `Assistant access via API` permission,
which is only available in your workspace with an additional purchase.
To get started, contact your Account Manager or reach out to us at
support@harvey.ai to discuss an agreement.
operationId: completion
parameters:
- name: include_citations
in: query
description: >-
Whether to include citations in the response output. When set to
false, citations are skipped and the response is returned more
quickly. Defaults to true.
required: false
schema:
type: boolean
default: true
example: true
requestBody:
required: true
content:
multipart/form-data:
schema:
type: object
required:
- prompt
properties:
prompt:
type: string
description: |
A string question that you want to ask Harvey.
The prompt may be up to 20,000 characters.
example: >-
In 3 sentences, can you explain how indemnity and limitation
of liability clauses typically work in commercial contracts?
default: >-
In 3 sentences, can you explain how indemnity and limitation
of liability clauses typically work in commercial contracts?
maxLength: 20000
stream:
type: boolean
description: >
A boolean value that lets you either stream or receive the
entire output at once.
default: false
file:
type: array
items:
type: string
format: binary
description: >
Files you want to upload to provide additional context for
the prompt.
Cannot be used together with `knowledge_sources`.
mode:
type: string
enum:
- draft
- assist
description: Optional mode to specify the behavior of the completion.
default: draft
example: draft
client_matter_id:
type: string
format: uuid
description: >-
Optional client matter identifier to associate with this
completion.
example: 12345678-1234-5678-1234-567812345678
knowledge_sources:
type: string
description: >-
JSON-encoded array of knowledge sources such as Vaults or
Regional Knowledge Sources.
Cannot be used together with `file`.
All objects must include a `type` field. Additional fields
depend on the `type`.
**Vault example with file_ids:**
[{"type":"vault","folder_id":"12345678-1234-5678-1234-567812345678","file_ids":["87654321-8765-4321-8765-432187654321"]}]
**Vault example without file_ids (queries entire folder):**
[{"type":"vault","folder_id":"12345678-1234-5678-1234-567812345678"}]
**Web example:**
[{"type":"web"}]
example: '[{"type":"web"}]'
model:
type: string
description: >-
Optional model selector. Pass a customer-facing model slug
to use a specific model for this request. Omit this
parameter to use your workspace's default model.
example: gpt-5
enum:
- gpt-5
- gpt-5-1
- gpt-5-2
- gpt-5-4
- gpt-5-4-mini
- gpt-5-5
- gpt-4-1
- claude-opus-4-5
- claude-sonnet-4-5
- claude-opus-4-6
- claude-opus-4-7
responses:
'200':
description: >
The response object includes the response field which is the
completion of the input prompt sent to the model.
content:
application/json:
schema:
type: object
properties:
response:
type: string
description: The completion of the input prompt.
example: >-
Indemnity clauses in commercial contracts require one
party to compensate the other for certain losses or
damages arising from specified events, such as breaches or
third-party claims. Limitation of liability clauses cap
the amount or types of damages one party can recover from
the other, often excluding indirect or consequential
losses. Together, these clauses allocate and manage risk
between the parties, providing predictability and
protecting against potentially catastrophic financial
exposure.
response_with_citations:
type: string
nullable: true
description: >-
The completion with inline citations, e.g. `[1][2]`, which
correspond to the sources below. Only returned when
`include_citations=true`.
example: >-
Indemnity clauses in commercial contracts require one
party to compensate the other for certain losses or
damages arising from specified events, such as breaches or
third-party claims.[1] Limitation of liability clauses cap
the amount or types of damages one party can recover from
the other, often excluding indirect or consequential
losses.[2]
sources:
type: array
nullable: true
description: >-
Contains the list of cited source snippets to support the
answer. Only returned when `include_citations=true` and
documents/knowledge sources were provided.
items:
type: object
properties:
citation_num:
type: integer
example: 1
document_name:
type: string
example: MyDocument.pdf
page:
type: integer
nullable: true
example: 2
text:
type: string
example: Quoted snippet...
text/event-stream:
schema:
type: string
description: >-
When `stream=true`, the endpoint returns Server-Sent Events
(SSE). Each event is a line starting with `data: ` followed by
a JSON object (e.g. `{ "response": "..." }`). The stream ends
with `data: [DONE]`.
example: >+
data: {"response":"Hello"}
data: {"response":" world"}
data: {"response":"","response_with_citations":"Hello
world.[1]","sources":[{"citation_num":1,"document_name":"MyDocument.pdf","page":2,"text":"Quoted
snippet..."}]}
data: [DONE]
'400':
description: Bad request.
'401':
description: Unauthorized.
'429':
description: Rate limit exceeded.
'500':
description: Internal server error.
security:
- bearerAuth: []
components:
securitySchemes:
bearerAuth:
type: http
scheme: bearer
bearerFormat: JWT
````