Skip to main content
PATCH
Update agent config

Authorizations

Authorization
string
header
required

Scopes

read
write

Path Parameters

projectKey
string
required

The project key

configKey
string
required

The agent config key

Body

application/json

Agent config object to update

defaultVariationKey
string

Key of the variation this config serves by default

description
string

A description of the agent config

maintainerId
string

The ID of the member who maintains this resource

Example:

"507f1f77bcf86cd799439011"

maintainerTeamKey
string

The key of the team that maintains this resource

Example:

"example-team-key"

name
string

A human-readable name for the agent config

tags
string[]

Tags for this resource

evaluationMetricKey
string

Evaluation metric key for this config

isInverted
boolean

The success direction of a judge. Offline evaluations use it to decide whether a score passes.

  • true: A lower score is better. Use this for a judge that scores bad output, such as a toxicity or hallucination judge.
  • false: A higher score is better. Use this for a judge that scores good output, such as an accuracy or relevance judge.

Send this field only when the agent config's mode is judge. If you omit it, the current direction does not change. For a built-in judge key, the API always returns the built-in direction. Built-in judge keys are ld-ai-judge-accuracy, ld-ai-judge-relevance, and ld-ai-judge-toxicity, with or without a -<timestamp> suffix such as ld-ai-judge-accuracy-1760000000000.

The API returns a 400 error in these cases:

  • isInverted is only supported for judge mode: You send this field and the agent config's mode is agent or completion. Remove the field.
  • isInverted is not supported: Custom judges are not available for your account. Remove the field from the request.
Example:

true

defaultThreshold
number<double>

The judge's threshold: the pass line for its main score, from 0 to 1. It applies wherever the judge is used. If you omit it, the current threshold does not change. To remove the threshold, send clearDefaultThreshold instead.

The API returns a 400 error in these cases:

  • Threshold is only supported for judges: You send this field and the agent config's mode is agent or completion.
  • Threshold must be from 0 to 1: The value is below 0 or above 1.
  • Thresholds don't support classifier rules yet: Every active variation of the judge uses classifier rules. Archived variations don't count.
  • Threshold can't be set and cleared in the same request: You send this field and clearDefaultThreshold: true.
  • Threshold is not supported: Thresholds are not available for your account. Remove the field from the request.
Example:

0.9

clearDefaultThreshold
boolean

Set to true to remove the judge's threshold. Don't send it with defaultThreshold.

The API returns a 400 error with Threshold is not supported when thresholds are not available for your account.

Example:

true

Response

Agent config updated

description
string
required

A description of the agent config

key
string
required

The agent config key

name
string
required

The agent config name

tags
string[]
required

Tags for this resource

version
integer
required

The version number of this resource

variations
object[]
required

The variations of this agent config

createdAt
integer<int64>
required
updatedAt
integer<int64>
required
_access
object
Example:

The location and content type of related resources

Example:
_site
object
Example:
_maintainer
object

The member or team that maintains this resource

Example:
mode
enum<string>
default:completion

The agent config mode

Available options:
agent,
completion,
judge
evaluationMetricKey
string

Evaluation metric key for this config

evaluationMetricKeys
string[]
deprecated

List of evaluation metric keys for this config

isInverted
boolean

The success direction of a judge. Offline evaluations use it to decide whether a score passes.

  • true: A lower score is better. Use this for a judge that scores bad output, such as a toxicity or hallucination judge.
  • false: A higher score is better. Use this for a judge that scores good output, such as an accuracy or relevance judge.

The response includes this field only when mode is judge. For a built-in judge key, the value is always the built-in direction. Built-in judge keys are ld-ai-judge-accuracy, ld-ai-judge-relevance, and ld-ai-judge-toxicity, with or without a -<timestamp> suffix such as ld-ai-judge-accuracy-1760000000000. A custom judge with no stored direction omits this field.

Example:

true

defaultThreshold
number<double>

The judge's threshold: the pass line for its main score, from 0 to 1. It applies wherever the judge is used. The average score passes when it is at or above this value, or at or below it when isInverted is true.

The response includes this field only when mode is judge and a threshold is set.

Example:

0.9

dependencies
object[]

Resources that depend on this config, grouped by type