Show HN: AgentRun – open-source, Self-improving Managed Agents (github.com)

AgentRun has launched as an open-source platform designed to enhance AI agents by enabling self-improvement, durable execution, and centralized control. This significant advancement allows agents to learn from past operations, storing useful notes, skills, and workflows to reduce redundancies and token usage in future runs. By incorporating capabilities like fault tolerance and real-time tracking through a single console, AgentRun optimizes the entire agent management process, making it more efficient and user-friendly. The platform supports various agent harnesses, including Claude Code, Codex, and OpenCode, allowing users to deploy agents in specific environments such as Docker containers or remote servers. Key features like approvals, spend caps, and a learning mechanism that evaluates and retains beneficial outcomes empower users to maintain control over agent operations. With a straightforward setup and integration capabilities, AgentRun presents a robust solution for developers looking to maximize the performance of their agents while minimizing operational costs. This innovation paves the way for smarter AI applications that evolve through their experiences, ultimately leading to more effective and economical AI-driven workflows.