Hi everyone,
I’m sharing Statefold, an open-source, framework-agnostic state and observability platform for AI agents.
Most agent systems store state as a mutable object, checkpoint or framework-specific memory format. That can make it difficult to understand exactly what happened during a run, safely resume after a failure, replay previous steps or reuse state across frameworks.
Statefold takes a different approach: agent state is derived by folding over an append-only event log.
For CrewAI applications, this can provide:
- Durable crash recovery and resume
- Time travel to any step in an agent run
- Replay of previous executions
- Branching for alternative or what-if runs
- Tracing of agents, tools and LLM calls
- Token, latency and cost tracking
- Working, semantic, episodic and procedural memory
- Tamper-evident, hash-chained event history
- PostgreSQL and in-memory backends
- A local observability console with a run waterfall and time-travel slider
Statefold is intended to sit underneath CrewAI rather than replace it. CrewAI continues to handle agent orchestration, while Statefold provides a durable and portable state layer.
Basic installation:
pip install "statefold[crewai]"
The project is Apache 2.0 licensed and currently in active development.
I would appreciate feedback from CrewAI users on:
- Which CrewAI state and memory workflows are hardest to debug today
- What information should be captured automatically from crews and tasks
- Whether replay and branching would be useful for testing multi-agent workflows
- What would be required before using this in production
Contributions, issues and integration feedback are welcome.