Skip to main content
POST
Valar does not keep conversation state. Every request is served from the content you send in input, so the fields that point at a previous turn held on the server carry no meaning here and are ignored rather than rejected: an item_reference input item, previous_response_id, and conversation. Your request still runs; it simply sees only what you sent.Send the conversation itself each turn — which every OpenAI-compatible SDK does by default — and set store: false so reasoning travels with it. See OpenAI API compatibility.

Authorizations

Authorization
string
header
required

Bearer authentication header of the form Bearer <token>, where <token> is your auth token.

Headers

Idempotency-Key
string

Lets you retry safely. Valar records a reservation under the combination of organization, API key, and Idempotency-Key, so sending the same value again hands back the stored response rather than running inference a second time. Values can be up to 255 characters. See Idempotent Requests for the complete rules.

Maximum string length: 255

Body

application/json
model
string
required
input
required

Accepts text, and image input (input_image) when you target a multimodal model. Valar does not yet accept audio, files, or item references.

Minimum string length: 1
max_output_tokens
integer | null
Required range: x >= 1
temperature
number | null
Required range: 0 <= x <= 2
top_p
number | null
Required range: 0 <= x <= 1
text
object
reasoning
object
background
boolean
prompt_cache_key
string

An optional hint Valar uses to keep prompt-prefix caches local. Requests sharing a key are steered toward the same place so you hit the cache more often.

store
boolean
default:true

How reasoning is returned to you. true (the default) is the stateful contract: your SDK will reference earlier reasoning by id, which Valar cannot resolve, so thinking does not survive a turn. Set false for an agent loop - reasoning then comes back with encrypted_content, the form SDKs replay, and Valar carries it to the model on your next request. This field does not control retention. See the Data Processing Addendum for retention commitments.

truncation
enum<string>
Available options:
disabled
stream
boolean
default:false

Set to true to receive Responses API Server-Sent Events instead of one JSON response. Streaming cannot be combined with background=true or the Flex completion window.

user
string
Maximum string length: 256
metadata
object

Optional string-valued metadata. Use completion_window to influence scheduling, and completion_webhook together with webhook_token to wire up a webhook that fires on completion.

Response

A JSON response, or a Server-Sent Events stream when stream=true.

id
string
required
object
enum<string>
required
Available options:
response
created_at
integer
required
status
enum<string>
required
Available options:
in_progress,
completed,
incomplete,
failed,
cancelled
model
string
required
usage
object
required
metadata
object
required
input

Accepts text, and image input (input_image) when you target a multimodal model. Valar does not yet accept audio, files, or item references.

Minimum string length: 1
output
error
object | null
incomplete_details
object | null
max_output_tokens
integer | null
reasoning
object
text
object
store
boolean
temperature
number
top_p
number
parallel_tool_calls
boolean
tool_choice
tools
object[]
truncation
user
string | null