Skip to main content
POST
/
api
/
v0
/
gepa
/
start_execution
Start GEPA Execution
curl --request POST \
  --url https://mainnet.dkn.dria.co/api/v0/gepa/start_execution \
  --header 'Content-Type: application/json' \
  --header 'x-api-key: <api-key>' \
  --data '{
  "customId": "<string>",
  "strategy": "RPM",
  "model": "gpt-5",
  "datasetColumns": [
    "feature1",
    "feature2"
  ],
  "budget": 100,
  "minibatchSize": 3,
  "paretoSize": 5,
  "evaluator": "{\"model\":\"gpt-5\",\"metric\":\"accuracy\",\"threshold\":0.8}",
  "dataset": [
    {
      "input": {
        "feature1": "value1",
        "feature2": "value2"
      },
      "expectedOutput": "The expected output for the given input features."
    }
  ],
  "prompt": "Write a poem about the sea."
}'
{
  "evolutionId": "<string>"
}

Authorizations

​
x-api-key
string
header
required

API Key

Body

application/json
​
strategy
enum<string>
required

The evolution strategy to use

Available options:
RPM,
SAM
Example:

"RPM"

​
model
string
required

The model to use for prompt completions

Example:

"gpt-5"

​
datasetColumns
string[]
required

The columns from dataset to use as input features

Minimum length: 1
Example:
["feature1", "feature2"]
​
budget
number
required

The total GEPA budget for the evolution session. Every inference request costs 1 budget unit.

Required range: x >= 1
Example:

100

​
minibatchSize
number
required

The size of minibatches.

Required range: x >= 1
Example:

3

​
paretoSize
number
required

The size of Pareto front to consider per iteration.

Required range: x >= 1
Example:

5

​
evaluator
string
required

Evaluator configuration as JSON string. Must include model

Example:

"{\"model\":\"gpt-5\",\"metric\":\"accuracy\",\"threshold\":0.8}"

​
dataset
object[]
required
Minimum length: 1
​
prompt
string
required

The root prompt to evolve

Minimum length: 1
Example:

"Write a poem about the sea."

​
customId
string

User given custom name for the evolution session, max 50 characters. If you do not provide one, a random name will be generated.

Maximum length: 50

Response

200 - application/json

Whether the execution was started successfully

​
evolutionId
string
required

Evolution ID