AI & ML

StateFabric.dev

Add durable, inspectable state to your AI agents

StateFabric is a small infrastructure layer for AI agents that need more context than chat history. Once an agent starts using tools, runs across multiple sessions, or needs to survive restarts, simply storing conversation history is no longer enough. You need to know: * What happened? * What state changed? * Which tools were called? * What context should go into the next model turn? StateFabric stores an append-only event log for every agent run, then derives working state and model context from those events. Rather than treating context as a growing message history, it treats it as a view built from durable events. Your agent framework or runtime continues to own prompts, tools, model calls, and business logic. StateFabric sits alongside it as the persistence, replay, and context layer. What it does StateFabric records every significant event that occurs during an agent session, including user messages, model responses, tool calls, tool results, and state updates. From that event stream it can: * Reconstruct session state after restarts. * Produce compact model-facing context instead of replaying entire conversations. * Provide a complete timeline for debugging and inspection. * Preserve a durable history for long-running agents. This approach makes it easier to build agents that run for extended periods, recover cleanly from failures, and maintain consistent context without relying on ever-growing chat histories. Who it’s for StateFabric is intended for developers building production AI agents that need persistence beyond a single chat session. Typical use cases include: * Long-running autonomous agents. * Multi-session assistants. * Tool-heavy workflows. * Agents that need reliable recovery after restarts. * Applications that require auditability or replay. * Custom agent runtimes where state management is currently built from scratch. It is not an agent framework. If you already use a framework, StateFabric is designed to complement it rather than replace it. Key features Current features include: * Durable session storage. * Append-only event log. * User, model, and tool event timelines. * Session reconstruction from stored events. * Compacted model-facing context. * Dashboard for inspecting sessions, raw payloads, compaction artefacts, and usage. * JavaScript integration for Google ADK via @statefabric/adk. * Direct Node.js and REST client via @statefabric/client for custom runtimes. The initial release deliberately focuses on a narrow set of integrations. Google ADK (JavaScript) and custom Node runtimes are supported today, with additional integrations planned over time. Pricing StateFabric is currently available as an early launch. Pricing is simple: * Free tier for evaluation and development. * Paid plans scale with usage as applications move into production. The goal is to provide infrastructure that can be adopted early in development without introducing unnecessary operational complexity. Quick answers (FAQ) Is StateFabric an agent framework? No. It does not manage prompts, tools, orchestration, or model calls. Your existing runtime remains responsible for agent behaviour. Does it replace LangGraph, Google ADK, OpenAI Agents SDK, or other frameworks? No. StateFabric is designed to sit beside agent frameworks as the persistence, replay, and context layer. Why an append-only event log? An immutable event history allows sessions to be replayed, state to be reconstructed deterministically, and debugging to be based on exactly what happened rather than a mutable snapshot. Why compact context instead of replaying chat history? Large message histories become increasingly expensive and often include information the model no longer needs. StateFabric derives a smaller working context from recorded events, helping keep prompts focused while preserving the complete history separately. What integrations are available today? The current release supports Google ADK (JavaScript) through @statefabric/adk and custom Node.js runtimes through @statefabric/client. Is this production ready? The project is in its early stages. The focus is on building a reliable foundation for durable state, replay, and context management before expanding integrations and capabilities.