02 / HARNESSES
Supported harnesses
Run Claude Code, Codex, OpenCode, Pi, Prime Agent, Hermes, DeepSeek Harness, Grok Build and Unreal Agent on Nebius models, each with a one-word NConnect command.
Launch any of them directly. Extra arguments are passed straight through to the underlying agent.
| Harness | Command | Mode | Notes | Tutorial |
|---|---|---|---|---|
| Claude Code | nclaude | Proxied | Anthropic Messages API translated to Nebius. | Guide |
 Codex CLI | ncodex | Proxied | OpenAI Responses API translated to Nebius. | Guide |
| OpenCode | nopencode | Spawned | Nebius wired in as an OpenAI-compatible provider. | Guide |
| Pi Code | npi | Spawned | Custom Nebius provider in a temporary config directory. | Guide |
| Prime Agent | nprime | Spawned | PrimeIntellect's RLM agent on Nebius models. | Guide |
 Hermes Agent | nhermes | Spawned | Nous Research's agent, isolated home overlay. | Guide |
| DeepSeek Harness | ndeepseek | Spawned | DeepSeek's web profile with Nebius layered in. | Guide |
| Grok Build | ngrok | Spawned | xAI's terminal harness; your key never reaches api.x.ai. | Guide |
 Unreal Agent | nunreal | Proxied | Unreal Labs' async-first runner; its Responses API traffic translated to Nebius. | Guide |
 Cursor | - | Editor | Native Nebius configuration. | - |
 Antigravity IDE | - | Editor | Native Nebius configuration. | - |
Proxied harnesses route through the local daemon. Spawned harnesses launch with a generated provider config. See How NConnect works.
Choosing a model
Put --model before the harness name. This works the same way for every harness:
nconnect --model moonshotai/Kimi-K3 hermesModel ids are listed on the Models page.