The fragmentation of the AI coding agent ecosystem just got a little less messy. Botiverse has released OAR (Open Agent Runtime), a provider-independent programming interface designed to unify how developers interact with various coding-agent runtimes. Instead of writing custom glue code for every new harness, OAR offers a solid foundation for building agent workspaces, allowing developers to delete harness logic and focus on outcomes and user experience.

One API to Rule Them All

OAR supports a wide array of runtimes, including Antigravity, Claude Code, Codex, Cursor, Grok Build, Kimi Code, OpenCode, Pi, Qwen Code, and Qoder. The core promise is simplicity: swap "claude" for "codex" or "grok" in your code, and the logic remains unchanged. If a runtime cannot honor a specific option, OAR refuses it before the session opens rather than silently dropping it, ensuring predictable behavior across different vendors.

Advanced Session Management and Subagents

The library provides granular control over sessions, including steering, queuing, and withdrawing prompts. It handles complex scenarios like handwork to other runtimes as subagents, where each report wakes the waiting session. OAR also introduces native goal support for compatible runtimes and provides utilities for installation, login, and account usage monitoring, all accessible via a single API or the accompanying CLI tool.

Technical Implementation

OAR is ESM-only and requires Node.js 24+ for its Node entry point, while the browser entry needs a modern browser with Web Crypto support. The package is licensed under Apache-2.0. It offers features like openVoyage for recording sessions to JSONL logs and @botiverse/oar/testing for scripted runtimes with real session semantics, eliminating the need for binaries or logins during development.

CLI and Documentation

The CLI exposes the same queries as the library, allowing users to check installations, login status, usage, models, and available skills via commands like oar installation and oar usage. Comprehensive documentation covers design decisions, the record stream spec, runtime mappings, and prior art comparisons, positioning OAR as a serious contender for the standard interface of AI coding agents.

Key Takeaways

  • OAR abstracts 10+ coding agent runtimes into a single interface.
  • Supports subagent orchestration and cross-runtime task delegation.
  • Includes a CLI for checking installations, usage, and upgrades.
  • Requires Node.js 24+ for the Node entry point.

The Bottom Line

This is the plumbing layer we've been waiting for. If you're building agent workflows, stop writing vendor-specific hacks and start using a standard.