Skip to main content
POST
Create experiment
Creates a new experiment with workflows. Supports three workflow types:
  • Custom: Submit your own workflow results via API
  • Completion: Direct LLM completions with custom parameters
  • Prompt: Load and render Jinja2 prompt templates with dataset variables
For custom workflows, the system creates placeholder traces that you update with your results. For built-in workflows (prompt/completion), execution starts automatically in the background.

Authentication

All endpoints require API key authentication:

Parameters

string
required
The name of the experiment.
string
Description of the experiment.
string
required
The ID of the dataset to run the experiment on.
array
required
List of workflow configurations.
string
required
Type of workflow. Options: custom, completion, or prompt.
  • custom: Submit your own workflow results
  • completion: Direct LLM completions
  • prompt: Load and render Jinja2 prompt templates
object
Configuration for the workflow. Structure depends on workflow type:
boolean
Allow trace updates (default: true).
number
Submission timeout in hours.
string
required
Model identifier (e.g., “gpt-4o-mini”).
number
Sampling temperature (0-2, default: 1.0).
integer
Maximum completion tokens (default: 150).
number
Nucleus sampling (0-1, default: 1.0).
number
Frequency penalty (-2 to 2, default: 0).
number
Presence penalty (-2 to 2, default: 0).
string or array
Stop sequences.
object
Response format (e.g., {"type": "json_object"}).
array
Function calling tools.
string or object
Tool choice strategy.
string
Reasoning effort for o1 models.
string
required
Prompt identifier to load and render.
array
List of evaluator slugs to run on the experiment results.

Response

Examples

Custom Workflow

Completion Workflow

Prompt Workflow

Workflow Rules

✅ Valid Combinations:
  • Single custom workflow
  • Single built-in workflow (prompt or completion)
  • Multiple built-in workflows chained together
❌ Invalid Combinations:
  • Multiple custom workflows
  • Custom workflow + built-in workflow
  • Mixing custom and built-in types
Chaining (Built-in Only): When you configure multiple built-in workflows, they execute in sequence - the output of one becomes the input of the next.