NEAR AI Gateway Connects Claude Code and Codex to Private Models

NEAR AI introduced NEAR AI Gateway, a macOS app that connects Claude Code, Codex and OpenCode to the private and incognito models on NEAR AI Cloud. The launch post, dated November 11, 2026, describes three jobs for the app: running coding agents against private models, analysing session traces locally, and paying NEAR AI Cloud credits for session data that users choose to share.

NEAR AI calls this release version 1 and says it is terminal-first. The pitch is aimed at developers who already pay for ChatGPT or Claude but sometimes want an idea kept out of training data, or a sensitive conversation handled by a private model, without switching tools.

01NEAR AI Gateway: what shipped in version 1

Gateway is a desktop app for macOS. From it, a user opens terminal sessions running Claude Code, Codex or OpenCode that are already configured to talk to NEAR AI Cloud instead of their default model providers.

The company frames the setup work as the main thing Gateway removes. Trying a private model, according to NEAR AI, should not mean reconfiguring each coding tool by hand, so the app handles that configuration and the user simply opens a session and starts working on a project.

Alongside the launcher, version 1 ships two other features: an Insights view built from local trace data, and an opt-in flow for contributing session data in exchange for NEAR AI Cloud credits. NEAR AI says the release is "only the start" and lists further work it has planned.

02Private and incognito models in Claude Code, Codex and OpenCode

NEAR AI Cloud offers two kinds of model, and the launch post is explicit that they protect data differently. Private models run in secure hardware enclaves that protect data while inference is happening.

Incognito models work another way. They reach external providers using NEAR AI Cloud's credentials, which means the user's own account is not attached, but those outside providers still process the requests. NEAR AI says its model catalog shows which protections apply to each model.

The company illustrates the use case with an unannounced product. An agent helping to build a prototype needs the code, the architecture, the business strategy and requirements still under debate, and NEAR AI positions private models as a way to hand over that context while keeping the work private.

The launch landed two days after NEAR AI's own security incident report, which said an SGLang flaw let an attacker observe requests on one inference server on November 8, including requests in confidential mode. NEAR AI says it has since deployed a fix across all inference servers and restored NEAR AI Cloud access.

03Session insights: token use by project and cost by model

The second feature turns agent sessions into something a user can review. NEAR AI argues that working with an AI agent involves repeated decisions, such as how to describe a task, what feedback to give, which approach to pursue and when a result is good enough, and that reviewing a session shows what made it productive or where it stalled.

Gateway lets users inspect trace data on their own Mac and generate insights from it. The Insights view breaks token usage down by project, so a user can see which project is consuming the most AI usage.

It also lists the files an agent kept re-reading, which NEAR AI describes as context the agent had already been given. Each session's cost is shown by model, and the company's argument is that knowing where tokens went makes the next session sharper.

04NEAR AI Cloud credits for opted-in session data

The third feature is a data exchange. NEAR AI says data from real development work reflects the problems people actually hit and the effort needed to reach a useful result, which gives it value beyond the project itself.

Developers who use AI on personal projects or open source contributions can contribute data from those sessions through Gateway and earn NEAR AI Cloud credits. The credits can be spent on future model usage.

Sharing is opt-in. NEAR AI says nothing leaves the machine unless a user opts in, traces are redacted on the Mac, and the user approves each trace before it is sent. The company expects people to mix modes, for example using confidential inference for an unreleased product while contributing data from an open source project.

05Gateway roadmap and macOS download

NEAR AI lists four areas it is working on after version 1:

The app is available now as a macOS download from NEAR AI Gateway. The launch post does not list Windows or Linux builds, and it does not give a credit rate for shared traces. NEAR AI asks early users to try Gateway on a real project and send feedback on what works and what does not.