Skip to content

Envoy AI Gateway becomes Agent Router and joins the Agentic AI Foundation

Learn more

Announcing Tetrate Desktop for Agent Router Enterprise

Tetrate Desktop finds the AI tools on your machine, configures them for Tetrate Agent Router Enterprise, and shows each tool's model, latency, and tokens.

Announcing Tetrate Desktop for Agent Router Enterprise

By John Landa, Product Architect, Tetrate

If you use Claude, Cursor, or similar AI tools, you already know the default setup: the tool holds an API key, picks a model, and calls the provider. That is fine on one machine. It is a poor fit when you want to see every request in one place or put those tools on a gateway your company already runs. People who are not engineers have a sharper version of the same problem. They have a tool they want to use, and no documented way onto the path platform and security teams will accept.

Today we are releasing Tetrate Desktop. Tetrate Desktop is a macOS app that finds Claude Code, Claude Desktop, and Codex on your laptop, routes them through a local Envoy AI Gateway to Tetrate Agent Router Enterprise, and shows each request’s tool, model, latency, and tokens. Install Desktop, sign in with your organization’s account, and Desktop configures your tools for you.

Why AI tools need a governed path to your company’s AI gateway

Every AI tool on a laptop talks to its model provider on its own terms. Each one has its own sign-in or API key, its own model list, and its own settings file. There is no shared point on the machine where an organization can see that traffic, or connect it to the AI gateway the organization already runs. That gap looks different for engineers and for everyone else.

Problems when AI coding tools talk to model providers directly

A developer might run Claude Code in one terminal, Codex in another, and Claude Desktop beside them. Each tool is configured separately, usually by editing a settings file or exporting environment variables, and each holds its own credential. That leads to three problems:

  • No attribution. Once requests leave the laptop, nobody can say which tool sent what, to which model, or at what token cost.
  • Setup per tool, per person. Moving a team onto the company’s AI gateway means walking each person through each tool’s configuration format. Base URLs and keys get copied into dotfiles by hand, and every copy can drift.
  • No local view. Developers have no record of what their own tools actually sent, how long each request took, or how many tokens it used.

Citizen builders have no path onto the company AI gateway

People outside engineering are adopting AI tools too. Someone in marketing or operations installs Claude Desktop because it helps them do their job. The security team is then left to choose between blocking it and looking away, because there is no documented install path that ends on the company’s governed gateway. Asking a non-engineer to edit a JSON configuration file is not a realistic answer.

What Tetrate Desktop does

Tetrate Desktop is the on-ramp from a laptop to Tetrate Agent Router Enterprise. You sign in with your organization’s account, Desktop finds the AI tools installed on your machine, and it configures them to send their requests through a local gateway that forwards them to your organization’s Agent Router Enterprise. It is more than a base URL and a reverse proxy: Desktop writes each tool’s settings for you, shows you the change before it makes it, and keeps a record so you can undo it.

Tetrate Desktop is a local AI gateway on your machine

Under the app is a real gateway, built on the open source Envoy AI Gateway (source on GitHub). It listens on http://127.0.0.1:1975, so only your own machine can reach it.

When you sign in, Desktop enrolls the install with Agent Router Enterprise. The key that enrollment creates is stored in the macOS Keychain and is never written into a tool’s settings. Each tool gets its own tool key instead. The gateway checks the tool key on every request and forwards the request to Agent Router Enterprise with the organization’s key.

Tetrate Desktop Tools page listing Claude Code, Claude Desktop, and opencode as connected, each with its own masked tool key, last-seen time, and requests in the last 24 hours
Tetrate Desktop Tools page listing Claude Code, Claude Desktop, and opencode as connected, each with its own masked tool key, last-seen time, and requests in the last 24 hours

Everything is inspectable. The desktop window and the headless tetrate serve daemon share one local control API, so anything you can do in the window you can also do from a terminal. Configuration lives as YAML files under ~/Library/Application Support/Tetrate Desktop/ that you can read, diff, and commit. Every change is staged with a diff before it applies, and any earlier configuration can be rolled back.

See which AI tool called which model, and what it cost

Because every tool has its own key, every request is attributable to the tool that sent it. The Overview shows traffic over time, along with a breakdown by model and by tool. The request log shows each request as it happens: the tool, the model it asked for and the model that served it, status, latency, time to first token, and tokens. Metrics adds an estimated cost, calculated at Agent Router Enterprise’s list prices.

Tetrate Desktop Overview with the gateway running, three tools connected, a traffic chart, and the Breakdown card by tool: claude-code 612, claude-desktop 214, and opencode 137 requests
Tetrate Desktop Overview with the gateway running, three tools connected, a traffic chart, and the Breakdown card by tool: claude-code 612, claude-desktop 214, and opencode 137 requests

Request and response bodies are captured locally, so you can inspect what a tool actually sent. They stay on your machine. Desktop sends nothing off the machine except the requests themselves, which go to your organization’s Agent Router Enterprise, where they are governed and recorded like any other traffic. If you want local metrics and traces elsewhere, you can turn on an OTLP exporter to your own collector.

You can also limit which models each tool may use, and change that from Desktop without touching the tool’s own configuration.

One Tetrate Desktop install for engineers and citizen builders

Engineers get a workbench: a CLI, plain YAML, curl access to the gateway, and the full request log. For how platform teams govern coding assistants at the gateway, see Enabling agentic developers with Tetrate Agent Router.

Everyone else gets a short path that needs no configuration files. Sign in, let Desktop scan for tools, tick the ones to connect, and confirm. For Claude Desktop, Desktop writes its third-party inference configuration for you and carries your local MCP servers into the new profile. IT teams can also pre-set the Agent Router Enterprise connection through a macOS configuration profile, so people only have to sign in.

It is the same app for both, and both paths end on Agent Router Enterprise.

Tetrate Desktop supports Claude Code, Claude Desktop, and Codex on Apple Silicon Macs

Who Tetrate Desktop is for

Tetrate Desktop is for organizations that run Tetrate Agent Router Enterprise and want every AI tool on every laptop going through it. Each person needs:

  • an account in your organization’s Agent Router Enterprise
  • at least one supported tool: Claude Code, Claude Desktop, or Codex

Desktop is a companion to Agent Router Enterprise, not a standalone gateway. It needs an Agent Router Enterprise sign-in before it connects any tool.

Onboard Claude Code, Claude Desktop, and Codex

Desktop sets up three tools automatically. For each one, it shows you the files it will change before it writes anything, keeps a record of the change, and lets you undo it later.

  • Claude Code. Desktop writes the gateway’s base URL and the tool’s key into ~/.claude/settings.json. That covers Claude Code in a terminal and the Claude Code that runs inside Claude Desktop.
  • Claude Desktop. Desktop adds a saved configuration to Claude Desktop’s third-party inference settings that points it at the gateway. No key is written to a file: a small helper hands Claude Desktop its tool key when it needs one. Claude Desktop runs gateway mode in its own profile, so you sign out of claude.ai, then quit and reopen Claude Desktop. Desktop copies your local MCP servers, from Claude Desktop and from Claude Code, into that profile, so your tools come with you. In this release, MCP traffic itself still goes directly from Claude Desktop to each server rather than through the gateway. Desktop automates the manual setup described in Adding Tetrate Agent Router as a third-party inference gateway in Claude Desktop.
  • Codex. Desktop adds the gateway as a model provider in ~/.codex/config.toml and puts the tool key in a marked block in your shell’s startup file, so a new terminal picks it up. Codex needs to use an API key rather than a ChatGPT sign-in to go through the gateway, which changes how its usage is billed.

Desktop also detects opencode. For opencode and any other tool that can send Anthropic Messages or OpenAI-compatible requests, Desktop creates a tool key and shows you what to paste into the tool’s settings.

Review screen for connecting Claude Desktop, with the What comes across notice listing the MCP servers copied into the gateway profile and what stays in the claude.ai profile
Review screen for connecting Claude Desktop, with the What comes across notice listing the MCP servers copied into the gateway profile and what stays in the claude.ai profile

What runs on your laptop vs in Tetrate Agent Router Enterprise

A request takes one path: your tool → Tetrate Desktop → Tetrate Agent Router Enterprise → the model provider.

On your laptop, Desktop owns:

  • the local gateway on 127.0.0.1:1975
  • one tool key per tool, and which models each tool may use
  • the request log, captured bodies, and metrics, all kept on your machine
  • the changes it made to each tool’s settings, and the record that lets you undo them

In Agent Router Enterprise, your organization owns:

  • the providers and models your tools can reach
  • policy and governance for that traffic
  • users, sign-in, and the key each Desktop install is issued at enrollment
  • the organization-wide view of usage across every install

How do you connect an AI tool to Tetrate Desktop?

Here is one path from start to finish: connecting Claude Desktop to Tetrate Agent Router Enterprise and seeing the first request.

First request with Claude Desktop through Tetrate Desktop

  1. Open Tetrate Desktop. It starts on “Set up Enterprise access.”

  2. Sign in. Enter your organization’s Agent Router API address and choose Continue. If your IT team pre-set the connection, Desktop skips this and opens directly on “Sign in to” your organization. Your browser opens on your organization’s sign-in page. Once you finish there, Desktop moves on by itself.

    Set up Enterprise access, step 1 of 4: the Agent Router API Address field and a Continue button
    Set up Enterprise access, step 1 of 4: the Agent Router API Address field and a Continue button

  3. Let Desktop prepare your access. Signing in enrolls this install: Agent Router Enterprise issues it a key, which Desktop keeps in the macOS Keychain and uses for every request it forwards. Desktop then starts the local gateway.

    Preparing your access, step 3 of 4: Tetrate Desktop minting this machine's inference key, registering the provider, and applying the configuration
    Preparing your access, step 3 of 4: Tetrate Desktop minting this machine's inference key, registering the provider, and applying the configuration

  4. Connect your tools. Desktop lists the AI tools it found on your machine. Tick Claude Desktop, choose to connect it, and review the files Desktop will change, including which MCP servers it will copy across. Confirm, and Desktop writes the files.

    Connect your tools, step 4 of 4: Claude Code and Claude Desktop selected, and Codex CLI unselected with a note that connecting it moves it off the ChatGPT sign-in
    Connect your tools, step 4 of 4: Claude Code and Claude Desktop selected, and Codex CLI unselected with a note that connecting it moves it off the ChatGPT sign-in

  5. Restart Claude Desktop. Sign out of claude.ai in Claude Desktop, then quit and reopen it. Its welcome screen says it is set up to run through a custom inference gateway. Choose Continue.

  6. Send a prompt. Pick a Claude model and ask something short.

  7. Look at the request. In Desktop, open Logs. The request is there with the tool (claude-desktop), the model, status, latency, and tokens. The Overview’s traffic and breakdown cards update too, and the same request is visible to your organization in Agent Router Enterprise.

    Logs screen listing requests from claude-code, claude-desktop, and opencode with the model, status, time to first token, total latency, and tokens for each
    Logs screen listing requests from claude-code, claude-desktop, and opencode with the model, status, time to first token, total latency, and tokens for each

The same steps work for Claude Code and Codex. Tick them in step 4. Claude Code picks up the change the next time it starts, and Codex picks it up in a new terminal.

Disconnecting from Agent Router Enterprise stops the gateway, removes the key from the machine, and asks Agent Router Enterprise to revoke it.

Getting started with Tetrate Desktop and Tetrate Agent Router Enterprise

Install Tetrate Desktop

This release runs on Macs with Apple Silicon.

  1. Get the latest .dmg and its SHA256SUMS file from your Tetrate team.

  2. Check the download against its published checksum:

    grep dmg SHA256SUMS | shasum -a 256 -c -
  3. Open the disk image and drag Tetrate Desktop to Applications.

  4. Open Tetrate Desktop. The app is signed with an Apple Developer ID and notarized by Apple, so macOS asks only its usual question for an app downloaded from the internet.

The local gateway starts as soon as you sign in to Agent Router Enterprise, as in the walkthrough above. On its first start, Desktop downloads the Envoy runtime once and checks it against digests built into the app.

If you prefer a terminal, the same app runs headless. tetrate serve starts the daemon without a window, and each step in the walkthrough has a CLI equivalent. Your Tetrate team can share the CLI reference.

For a team rollout, IT can pre-set the connection through MDM, so people skip the address step and go straight to signing in. Deploy a macOS configuration profile for the io.tetrate.desktop domain, or a root-owned managed.json file at /Library/Application Support/Tetrate Desktop/. It sets the Agent Router Enterprise addresses for everyone. Desktop shows them as “Set by your organization” and opens on the sign-in. The same settings can turn on start at login and leave updates to IT.

What’s next for Tetrate Desktop

We are always adding support for more AI tools and the platforms people run them on. Next up:

  • More tools. Cursor and Coder support, along with further improvements for Codex.
  • More platforms. Intel Macs, Windows, and Linux.

Learn more on the Tetrate Desktop page. If your organization already runs Tetrate Agent Router Enterprise, ask your Tetrate team for the install. To see Desktop with your tools, book a demo.

Frequently asked questions about Tetrate Desktop

What is Tetrate Desktop?

Tetrate Desktop is a macOS app that connects the AI tools on a laptop to Tetrate Agent Router Enterprise. It finds Claude Code, Claude Desktop, and Codex, configures each one to send requests through a local Envoy AI Gateway on 127.0.0.1:1975, and logs each request’s tool, model, latency, and tokens.

Which AI tools does Tetrate Desktop support?

Tetrate Desktop sets up Claude Code, Claude Desktop, and Codex automatically. It also detects opencode. For opencode and any other tool that sends Anthropic Messages or OpenAI-compatible requests, Desktop creates a tool key and shows the settings to paste into the tool. Cursor and Coder are not supported in this release and are planned next.

Does Tetrate Desktop work without Tetrate Agent Router Enterprise?

No. Tetrate Desktop is a companion to Tetrate Agent Router Enterprise, not a standalone gateway. You must sign in with your organization’s Agent Router Enterprise account before Desktop connects any tool.

What operating systems does Tetrate Desktop run on?

This release runs on Macs with Apple Silicon. Support for Intel Macs, Windows, and Linux is planned.

Where does Tetrate Desktop send my data?

Request and response bodies, the request log, and metrics stay on your machine. The only traffic that leaves the machine is the model requests themselves, which go to your organization’s Agent Router Enterprise. You can turn on an OTLP exporter to send local metrics and traces to your own collector.

Does Tetrate Desktop route MCP traffic?

Not in this release. When Desktop connects Claude Desktop, it copies your local MCP server settings from Claude Desktop and Claude Code into the gateway profile. MCP traffic still goes directly from Claude Desktop to each MCP server.

Do I have to sign out of claude.ai to use Claude Desktop with Tetrate Desktop?

Yes. Claude Desktop runs gateway mode in its own profile. Sign out of claude.ai in Claude Desktop, then quit and reopen Claude Desktop. Its welcome screen then says it is set up to run through a custom inference gateway.

Does connecting Codex change how Codex usage is billed?

Yes. To go through the gateway, Codex must use an API key instead of a ChatGPT sign-in, so Codex usage is billed through the API key path instead of a ChatGPT plan.

How do IT teams deploy Tetrate Desktop with MDM?

IT can pre-set the Agent Router Enterprise connection with a macOS configuration profile for the io.tetrate.desktop domain, or with a root-owned managed.json file at /Library/Application Support/Tetrate Desktop/. Desktop then shows the addresses as “Set by your organization” and opens on the sign-in screen. The same settings can turn on start at login and leave updates to IT.

Product background Product background for tablets
Building AI agents

Agent Router Enterprise provides a managed AI Gateway, MCP Gateway, and AI Guardrails in your dedicated instance. Graduate agents from prototype to production with consistent model access, governed tool use, and runtime supervision — built on Envoy AI Gateway by its creators.

  • AI Gateway – Unified model catalog with automatic fallback across providers
  • MCP Gateway – Curated tool access with per-profile authentication and filtering
  • AI Guardrails – Enforce policies, prevent data loss, and supervise agent behavior
  • Learn more
    Replacing NGINX Ingress

    Tetrate Enterprise Gateway for Envoy (TEG) is the enterprise-ready replacement for NGINX Ingress Controller. Built on Envoy Gateway and the Kubernetes Gateway API, TEG delivers advanced traffic management, security, and observability without vendor lock-in.

  • 100% upstream Envoy Gateway – CVE-protected builds
  • Kubernetes Gateway API native – Modern, portable, and extensible ingress
  • Enterprise-grade support – 24/7 production support from Envoy experts
  • Learn more
    Decorative CTA background pattern background background
    Tetrate logo in the CTA section Tetrate logo in the CTA section for mobile

    Ready to enhance your
    network

    with more
    intelligence?