Agent Dev Loop

BASE CONFIG

Base Config

Config Name

Edit the middle part. Prefix (parent config) and suffix (auto-increment) are fixed.

MODEL SETTINGS

Model

Temperature

DATASET

Dataset Path

Select from available datasets in datasets/ directory

Dataset Preview

Select a base config to view dataset fields

💡 Use field names like {problem_description} in messages below

INITIAL MESSAGES

ENVIRONMENT SETTINGS

SSH Target

GPU Ranks

DATASET SETTINGS

Seed

Sample Range

0 10

Max Turns

Unlimited

Use -1 for unlimited turns

Stream Tokens

Show live LLM token streaming in the output (recommended for dev loop)

TOOLS

Select a base config to view and edit tool descriptions

ENVIRONMENT

Environment Source Code

View the full environment implementation (prepare_messages, get_tools, on_assistant_message, etc.)

Welcome to Agent Dev Loop

Select a result to view or configure and launch a new agent evaluation.