Skip to main content
The Nito plugin lets you call open-weight models like Kimi K3, GLM 5.2, DeepSeek, and Qwen, plus frontier models like GPT, Claude, and Gemini, from inside Claude Code and Codex. Your harness keeps running its own model. Ordinary messages never touch Nito, and a call goes through Nito only on the commands you type. Nothing in your existing setup changes. Every Nito call runs at a named privacy level: anonymous, private, confidential, encrypted (coming soon). The privacy level comes with the model you select. Nito-routed frontier models always run anonymously, so the provider gets your prompt from Nito without your identity attached. Open-weight models go further: from zero data retention to running inside attested hardware that hands back proof of privacy and model usage. In all cases, Nito keeps neither the prompt nor the response. Nito commands come as skills: in Claude Code they are /nito:…; in Codex they are $nito:…. Everything else in your session works exactly as before.

​
What You Can Do

Use any model

/nito:ask sends your prompt to whichever model you pick from the Nito catalog and puts the response in your session. Name a model per call with --model, or set a default once and leave it alone.

Prompt privately

Type /nito:private with any prompt to call a model in the Nito catalog, incognito. The response lands in your session without your harness provider having any record of the prompt or response.

Enter full Nito sessions

/nito:session opens a fresh session in your harness where you can access and switch between all models in the Nito catalog, with your history retained for future sessions. Enter /nito:private with no prompt for incognito sessions: full conversations without prompts or responses being recorded upstream.

Deploy multiple models on one prompt

/nito:fusion sends one prompt to two or three models, then returns a synthesized, stronger response.

​
Same Harness. More Models.

The plugin works in Claude Code and Codex rather than replacing either or requiring migration. Same harness, same configuration, same tools. It does not swap out the model running your session, rewrite your settings, or change your ordinary prompts. When you run a Nito command, it runs through Nito. When you don’t, nothing does.

​
Privacy Levels

Models come with a designated privacy level. z-ai/glm-5.2 runs at Private; z-ai/glm-5.2:confidential runs at Confidential.

Anonymous

Frontier models run anonymously with your prompt routed through Nito so your identity is unattached.

Private

Open-weight models with zero data retention upstream, so the provider does not store your prompt or the response.

Confidential

Open-weight models run privately inside attested hardware. Nito authenticates the attestation before you see the answer, and blocks the response if the check fails.

Encrypted (coming soon)

Your prompt is encrypted on your own machine against the hardware’s key, so nothing in between can read it. Web search, file uploads, and tool calls are switched off at this level.
A privacy level never silently weakens. If you ask for a stronger level than a model supports, the call is refused with a clear message rather than served at a lower privacy level. The full ladder is covered in Privacy.

​
You Choose What Your Harness Provider Sees

/nito:ask is an ordinary turn in your session, so it travels upstream with the rest of your conversation like everything else you type. /nito:private doesn’t. Your prompt runs through Nito instead, the response returns incognito inside your session, and your harness never receives either upstream. Nito keeps no copy. Run your first private prompt.

​
Free vs Paid

You can start on a free allowance and upgrade when you need more. Full details, including what counts against the free allowance, live in Free Usage.

​
One Plugin, Two Hosts

Nito can live inside both Claude Code and Codex. The same commands exist in both; only the trigger differs (/nito: in Claude Code, $nito: in Codex).

Install in Claude Code

Install, authorize, and run your first ask.

Install in Codex

Install, authorize, and run your first ask.