DeepYardDeepYard
D

DeltaBox

Millisecond-level checkpoint/rollback system for stateful AI agent exploration

Open SourceFree

About

Research sandbox system enabling rapid state exploration for AI agents through ultra-fast checkpoint and rollback operations. Achieves 100-1000x speedup over traditional methods, making it practical for test-time search, reinforcement learning training, and agent tree search scenarios. Designed for researchers and developers building agents that require frequent state branching and backtracking.

Details

Type
Integrations
Language

Tags

autonomousopen-sourceframeworkevaluationpythonself-hosted