smithers
by smithersai
Durable, restartable runs for coding agents with pause, rewind, and human approvals
Summary
Manages durable runs for coding agents so workflows survive crashes, interruptions, and human review. Persists execution state, supports pausing for human approval, and enables rewinding or retrying mistaken steps across models like Claude Code, Codex, and Gemini. Ships as a JavaScript/TypeScript library with background agent support and developer-friendly hooks for web UIs. See the LLM-as-Judge Pattern for related architectural guidance.
Key Benefits
Ideal For
Teams building production coding agents who need long-lived, recoverable workflows with human-in-the-loop controls and auditability. It complements governance workflows by integrating with the Model Context Protocol (MCP) to standardize context and evaluation across agents.
Use Cases
- Recovering long-running code-generation tasks after crashes or restarts
- Pausing agent workflows for human approval or code review before deployment
- Rewinding or retrying agent steps to fix logic errors or hallucinations
- Running background agents that need persisted state and observability