**The stateful control plane for AI agents.** Crash on step 47, resume from step 46. Budget the whole run at $2. Kill rogue agents instantly. All local, no SaaS, no framework lock-in. Works with multiple LLM SDKs (Anthropic, Gemini, OpenAI, Cohere, Mistral, LiteLLM, etc.) and agent CLIs like Claude Code, Gemini CLI, and [OpenClaw](OPENCLAW.md). --- ## Why StateLoom? StateLoom is **session-aware**. Instead of seeing each API request in isolation, it sits in the request path and groups fragmented multi-step workflows into meaningful, stateful sessions. This lets you resume crashed scripts without re-paying for completed steps, enforce budgets across an entire agent run (not just a single call), contain blast radius when things go wrong, and get full visibility into what your models are actually doing. It runs locally on your laptop or inside your VPC. Prompts never leave your network. --- ## Table of Contents - [Why StateLoom?](#why-stateloom) - [Quick Start](#quick-start) - [Providers](#providers) - [Core Features](#core-features) - [Advanced Features](#advanced-features) - [More Features](#more-features) - [Key Examples](#key-examples) - [Dashboard](#dashboard) - [For Teams & Enterprise](#for-teams--enterprise) - [Extras](#extras) - [Configuration](#configuration) - [Error Handling](#error-handling) - [Documentation](#documentation) - [Contributing](#contributing)

Agents

stateloom

context-chef