Statefold

Statefold – Event-sourced state, replay, timetravel

12 followers

Statefold – Event-sourced state, replay, timetravel

12 followers

Visit website

Framework-agnostic, event-sourced state platform for AI agents (LangGraph, CrewAI, Agno, MCP) - ioteverythin/statefold

Free

Launch tags:Open Source•Artificial Intelligence•GitHub

Launch Team

Customer.ioAutomate Messaging Everywhere — Startups Get 12 Months Free

Promoted

Maker

📌

We built Statefold because agent frameworks tend to treat state as a mutable object, checkpoint, or framework-specific memory store. That makes it difficult to inspect exactly what happened, resume safely after a failure, replay a run, or move state between frameworks. Statefold models agent state as a fold over an append-only event log. Each session records messages, state changes, LLM calls, tool calls, traces, usage, costs, and memories as events. The current state is derived by replaying those events, while checkpoints are only an optimization. This provides: Durable crash recovery and resume Time-travel to any point in a run Replay and O(1) branching for alternative runs Optimistic concurrency with expected sequence numbers Hash-chained, tamper-evident event history Tracing and cost observability from the same event log Working, semantic, episodic, and procedural memory Export of production conversations into Promptfoo evals A local event-log inspector with a time-travel slider In-memory and PostgreSQL backends Integrations for LangGraph, CrewAI, Agno, raw SDK clients, and MCP The core package has no required dependencies and is licensed under Apache 2.0. The main design question I am exploring is whether event sourcing can serve as a neutral state layer beneath different agent frameworks, instead of every framework maintaining its own incompatible state model. We would especially appreciate feedback on the event model, reducer API, framework integrations, and what would be required before using something like this in production

Report

2mo ago

A timeline view for replaying agent state changes would be incredibly useful for debugging - being able to scrub through event history visually and see exactly when context was lost or corrupted. Could really speed up troubleshooting when agents start behaving weirdly.

Report

2mo ago

a visual replay debugger would be huge here, basically something where you can scrub through the event log and see the agent's state at each step. would make debugging way less painful than digging through raw event streams

Report

2mo ago

finally something that works with my existing langgraph setup without me rewriting half my codebase, basically plug and play

Report

2mo ago

finally something that lets me swap agent frameworks without rewriting state logic. tried it briefly with langgraph and the event replay actually worked first try.

Report

2mo ago

love that it stays framework-agnostic instead of locking you into one stack, that's the kind of decision that shows they actually thought through how people build these days

Report

2mo ago

Reviews