Post-train your model on your own agent harness with OpenHands, Harbor, SkyRL and Snowflake
StasBekman · x · 2026-11-04
Karthik Ganesan details a stack that lets product companies post-train models on their own agent harnesses — the way frontier labs do (e.g., xAI says Grok 4.7 was trained to understand the Grok Bot harness).
Component roles:
- OpenHands: customizable agent harness plus a skill that sets up and runs training
- Harbor: tasks, environments and tests providing the reward signal
- SkyRL: the RL training loop, integrated with Arctic RL
- Snowflake Cortex Training: managed GPUs for training/sampling, with isolated Cortex Sandboxes for rollouts
- Arctic Platform: backend recipes connecting the pieces
Once Harbor tasks and training access are ready, you send your OpenHands agent one command — it launches training, tracks rewards and failed rollouts, plots progress, and releases GPUs when done. Your prompts, tools, skills and turn budgets carry straight into training without rebuilding the agent in a separate RL framework.
More from coding & agent
- Revolut Moves 2,000 Engineers to Open Models for Coding: 1/10th Cost, 70% Faster Lead Time — zainhas · 2026-11-04
- Assigning Agents Standard/High/Max Effort Levels Is Just Story Points for Machines — feregri_no · 2026-11-04
- 博主晒海外 AI 数字员工分工：GPT 管后端、Claude 剪片、Gemini 写文案 — lxfater · 2026-11-04
- Aixio 推出托管 MCP 服务，让编码 Agent 直连图像视频设计工具 — aiiiiiaiaiaiaiwater · 2026-11-04
- Opsec runs a fleet of parallel AI security agents that find exploits and auto-open fix PRs — AccBalanced · 2026-11-04
- Teknium clarifies Hermes Agent works with any subscription, cloud or otherwise — Teknium · 2026-11-04