OpenEnv to add multi-harness RL training across Claude Code, Codex, Gemini CLI and more
SergioPaniego · x · 2026-09-17
Developer adithyask previewed upcoming multi-harness RL training in the fully open-source OpenEnv project.
- The pitch: "pick a model, pick a harness, pick a sandbox, train" — decoupling RL environments from coding agents.
- Supported harnesses at launch include Claude Code, Codex, Gemini CLI, OpenCode, Pi, Kimi, and OpenHands, among others.
- It uses an async RL loop, is end-to-end Harbor compatible, and is dropping soon.
More from coding & agent
- PromptingBox Launches MCP Server for Prompt Management in Claude and Cursor — modelcontextprotocol · 2026-09-17
- REG-Vault Open-Sources Retro Gaming Metadata: 91,000 Games via Free MCP API — modelcontextprotocol · 2026-09-17
- A Trading Bot Built With Jev Places Real On-Chain Trades Every 300ms Block — cyrus_zei · 2026-09-17
- Embeddings Handbook Follow-Up: The Author Shares the Full Read Link — techNmak · 2026-09-17
- An Embeddings Handbook: Why Token, Contextual and Sentence Embeddings Are Not Interchangeable — techNmak · 2026-09-17
- Android Use debuts: a scheduled, memory-equipped Android agent with permission controls — emeka_boris · 2026-09-17