Checkpointing Python Packages

substrax

JAX/Flax NNX training infrastructure: device meshes and SPMD sharding, an Orbax checkpoint store, early stopping and callbacks, W&B and MLflow logging.

datarax

Differentiable data pipelines for JAX/Flax NNX: HuggingFace, TFDS and ArrayRecord sources, augmentation stages, scan-based epochs, exact mid-epoch resume.

sayiir

Sayiir: simple, embeddable durable workflow engine in Rust with Node.js & Python bindings. Checkpoint-based recovery, no deterministic replay. A simplified alternative to Temporal, Restate, and Airflow.

async-batch-llm

Provider-agnostic framework for high-throughput LLM processing with async workers, automatic retries, rate limiting, and intelligent validation recovery.

pyattacker

Resumable async task orchestration for Python — pipelines, artifact persistence, per-endpoint pools. One failure costs a task, not the run.

context-compiler

Deterministic state engine for managing conversation state and constraints in LLM applications.

trigora-cli

High-performance, lightweight durable execution for long-running AI agents and programs.

ckpt-chaos

Crash-consistency testing for ML training checkpoints: kill a save at every file boundary, resume, classify the outcome

context-compiler-example-integrations

Example integrations and enforcement-point demonstrations for Context Compiler.

chidori

The agent framework where every run is durable, replayable, and resumable by default.

torchgpipe

A GPipe implementation in PyTorch

trigora-client

Python SDK and client for Trigora durable execution.

trigora

Python SDK and client for Trigora durable execution.

dagpipe-core

Zero-cost, crash-proof LLM pipeline orchestrator. Features disk-based checkpointing, free-tier routing, and structured output. (LangGraph / CrewAI alternative)

trainparity

A lightweight PyTorch library for verifying training parity across resume, gradient accumulation, and sample coverage, with first-divergence localization.

dlio-benchmark

An I/O benchmark for deep Learning applications

saveable-objects

A Python package for checkpointing, saving, and loading objects.

loomc

loomc is a Python decorator that turns a single-iteration function into a resilient agent loop — with hard budget caps, SQLite checkpointing after every step, and full resume from wherever it stopped.

thread-chunks

A Python package for performing memory-intensive computations in parallel using chunks and checkpointing.

sylo-sdk

Sylo is an open-source Python SDK and cloud platform that adds production-grade reliability to your existing AI agent pipelines — checkpointing, permission enforcement, human approval gates, and immutable audit logging — without replacing the frameworks you already use.

continuum-agent

CONTINUUM: Verifiable semantic recovery for long-running AI agents. Semantic checkpoints (not conversation dumps), an idempotent action ledger that refuses duplicate side effects, and a hash-chained tamper-evident event log, all exposed as a deny-by-default MCP server. Framework-agnostic, Python 3.11+.

chronalyn

Long-term memory for Hermes Agent, with Hindsight recall and verified Mnemosyne checkpoints.

agent-recall-ai

Your AI agent never starts over again. Framework-agnostic session checkpointing - survive context limits, cost overruns, and session death.

agentstateprotocol

Checkpointing and recovery protocol for AI agents