YiYaoAI

Posted on Apr 10

"CliGate Now Has a Built-in AI Assistant That Can Configure Your Proxy For You"

#ai #productivity #devtools #opensource

Most local dev tools give you a config file and a README. If something breaks, you're on your own.

CliGate just shipped something different: a built-in AI assistant that lives inside the dashboard, understands the product, and can actually do things for you.

What Is CliGate Again?

CliGate is an open-source local proxy that sits between your AI coding tools and their APIs. You point Claude Code, Codex CLI, Gemini CLI, and OpenClaw at localhost:8081 — and CliGate handles routing, account pooling, protocol translation, and failover.

npx cligate@latest start

The dashboard opens at http://localhost:8081.

The New Chat Page

There's now a Chat tab in the dashboard.

On the surface it looks like a chat interface — and it is. You pick a credential source (a ChatGPT account, Claude account, or any API key you've added), choose a model, optionally set a system prompt, and start chatting. It's a useful testing surface for verifying that your credentials actually work before routing real CLI traffic through them.

But that's the boring part.

The Product Assistant Mode

Here's where it gets interesting.

Toggle on Product Assistant, and the chat behavior changes.

The assistant now has the full CliGate product manual loaded into its context. Ask it things like:

"How do I configure Codex CLI to use my Azure key?"
"What's the difference between Account Pool First and API Key First routing?"
"How do I enable free model routing?"

And it answers with actual, accurate information about this specific product — not generic AI hand-waving.

This is useful. CliGate has a lot of moving parts: multiple protocols, multiple account types, routing modes, model mapping, Gemini patching, free model fallback. Having an assistant that knows the system well enough to answer specific setup questions in plain language removes a lot of friction for new users.

The Action Mode: Chat That Does Things

This is the part that surprised me.

Type something like:

"Set up Claude Code to use the proxy"

And the assistant doesn't just tell you how. It shows you a confirmation card:

Enable Claude Code Proxy
Configure Claude Code to use the local proxy at http://localhost:8081.
[ Confirm ]

Click Confirm, and it actually writes the configuration to your Claude Code credentials — switching it to proxy mode, pointing it at localhost:8081, and mapping the model aliases.

Same in reverse: ask it to disable the proxy, and it confirms before restoring direct mode.

The token-based confirm step isn't just UX polish. It's a deliberate safety gate. The action token expires in 10 minutes. Nothing changes without your explicit confirmation. The assistant proposes, you approve, the action executes.

Why This Matters More Than It Looks

Most AI tools have chat interfaces.

Very few of them are product-aware assistants that can actually modify their own configuration on your behalf.

The gap between "I know how to fix this" and "I have just fixed this" is where most tool friction lives. CliGate's assistant collapses that gap for the most common setup operations — at least for Claude Code proxy toggle right now, with more actions likely coming.

The language support is also worth noting: the assistant detects whether you're asking in English or Chinese and responds accordingly. The intent detection and tool pattern matching work across both languages.

The Practical Loop

Here's what the workflow looks like now for a new user:

npx cligate@latest start
Open http://localhost:8081
Add an account or API key in the Accounts / API Keys tab
Go to Chat → enable Product Assistant
Ask: "How do I set up Claude Code to use this proxy?"
The assistant explains it, then offers to do it for you
Click Confirm
Done

That's a pretty clean onboarding path for what used to require navigating Settings, reading docs, and manually editing config files.

DEV Community