Tip Of AI
Discover what you can do with AI
AgentRun is an open source platform for managing AI agents across multiple agent harnesses including Pi Claude Code Codex and OpenCode. It provides one API and console for defining agents starting runs managing revisions inspecting results and responding to requests for input. The platform adds self improvement by turning repeated steps into proposed skills notes and tested workflows that later runs can reuse. This helps reduce repeated planning tool discovery and reasoning although the website states that cost and token improvements are illustrative rather than measured benchmarks. AgentRun also supports durable execution by saving conversations files completed tool results and run state. Interrupted runs can continue from saved context while crash recovery can reuse recorded results for the same tool call. Each run can have its own files and can connect to tools and remote MCP servers with explicit access controls. Compute options include Docker Daytona and Modal with lightweight development and high resource sandbox configurations. AgentRun is available for self hosting under the Apache 2.0 license or through AgentRun Cloud and enterprise deployment options. Users bring their own model API keys and manage infrastructure costs when self hosting. Client libraries are provided for TypeScript and Python.