Works with 12 AI coding clients

One switch,
switch your AI coding clients

Q: What does AiSwitch do?

It is a small Windows and macOS desktop app that switches AI coding clients between their official endpoints and third-party APIs in one click. It handles config files, environment variables and model mapping for you.

Q: Which clients are supported?

Twelve today: ChatGPT (Codex), Claude Code, Claude Desktop, Antigravity, Grok Build, OpenCode, OpenClaw, MiniMax Code, ZCode, plus WorkBuddy, WorkBuddy China and WorkBuddy Global. More come with each release.

Q: Will switching lose my existing config?

No. The app backs up the original config before every switch, and "Switch to official" restores it as it was. The backup is a plain file you can open yourself.

Q: Are my API keys uploaded anywhere?

No. Reading and writing config, switching and latency checks all happen on your machine. There is no account system and no relay through our servers.

Q: Is it free? Do I need an account?

Free, no account. On Windows it needs no admin rights — install it and open it.

Q: How is this different from editing config.toml by hand?

Hand-editing means remembering a different path and schema for every client, every time. AiSwitch keeps that in one window and also handles model catalog injection and reasoning mapping, which are easy to get wrong.

Q: Can I switch back to the official endpoint?

Yes. Click "Switch to official" and it restores the original config and cleans up the temporary files it wrote. The client picks it up after a restart.

Q: Can one API endpoint serve several clients?

Yes. You keep it once under "My API" and every client shares the same endpoint and key, so a change propagates everywhere.

Move between official endpoints and third-party APIs in one click.
No more hand-editing config.toml, no more empty balances or missing models.

Download AiSwitch See what it does

Windows 10+ · macOS 12+ · 10.1 MB installer · No admin rights

AiSwitch

ChatGPTCodex Managed Search channels New channel

ChannelEndpoint & keyDefault modelLatencyAction

Official Official endpointNot managed gpt-5.4-codex — Use it

Provider A api.example.com/v1sk-••••••••7f3a gpt-5.4-codex 128ms In use

Provider B relay.example.net/v1sk-••••••••2c19 claude-sonnet-5-5 342ms Use it

Local gateway 127.0.0.1:8317/v1sk-local-•••• auto 12ms Use it

clients

platforms

per switch

config edits

You only wanted a new channel

The whole evening went into config files.

One wrong character, blank screen

Every client has its own format. TOML, JSON, all different. A missing comma or an extra space and it will not start, with no hint where it broke.

New provider, full rewrite

Rate limits, price hikes, a provider that disappears. You switch anyway. Each time, five clients get reconfigured from scratch.

Missing from the model list

Third-party models lack fields the client expects, so they never show up, or quietly fall back to a default. You pay for one model and run another.

Core features

All the switching, in one window

Official or relay, one click apart

Pick a channel and press Switch. AiSwitch takes over that client's config and restarts it. Going back to official is the same click.

✓Drag channel cards to reorder, keep frequent ones on top
✓Live latency checks, dead channels show up at a glance
✓Endpoint and key on separate lines, each copied on its own

Channels · ChatGPT

Provider A api.example.com/v1 128ms In use

Provider B relay.example.net/v1 342ms Use it

Official Default channel Back to official

Models not showing? Map them once

Relay models usually lack context window and reasoning fields, so official clients cannot read them. AiSwitch maps each one and injects a catalog the client accepts.

✓Model ID, context window and reasoning level on one row
✓Pick the real model from a dropdown, no names to memorize
✓Set it as default and it applies on the next switch

Model mapping · per channel

Name	Real model	Context	Reasoning
Flagship	gpt-5.x-codex	400K	High
Fast	gpt-5.x-mini	200K	Mid
Long docs	long-context-v2	1M	Low

Fields are filled in on switch, no fallback to defaults

One credential, every client

Enter the endpoint and key once in My API. Reference that entry from any client. Edit once and every client follows.

✓Credentials and model mapping stored apart, a new provider needs no re-entry
✓Add one credential to several clients in one click
✓Keys stay local and are masked in the UI by default

My APIs

Relay A · primary sk-••••••••4f2a

Relay B · backup sk-••••••••91c7

Local gateway sk-••••••••0d33

Claude Desktop with third-party models

The desktop app blocks non-Anthropic model names outright. AiSwitch runs a local reverse proxy that exposes only compliant names, rewrites them to the real model upstream, and keeps streaming output intact.

✓Compliant names outside, real models restored upstream
✓SSE passes straight through, token by token output intact
✓Switching back cleans up leftover config

Request path

Claude Desktop → Local proxy 127.0.0.1:15888 → Relay / official

Exposed as claude-sonnet-4-5

Real model deepseek-v4 / kimi-k2 / glm-5

Clients

Most of what you already run is here

Every client is taken over differently. AiSwitch adapts to each one.

ChatGPT · Codex

Model catalog injection and reasoning depth

Claude Code

Environment and config injected in one click

Claude Desktop

Local reverse proxy with compliant renaming

Antigravity

Language server wrapper, multimodal passthrough

Grok Build

Channel switching for terminal agents

OpenCode

Config wiring for an open-source coding agent

OpenClaw

Channel switching for agent orchestration

MiniMax Code

Seamless native config switching

ZCode

Zhipu channels and custom endpoints

DeepSeek Harness

Model switching for agent environments

WorkBuddy CN

Official Hunyuan channel access

WorkBuddy Global

International channel access

Security and privacy

Your keys never leave this machine

AiSwitch is a local switch. It does not proxy your requests or touch your conversations.

All local

Config reads and writes, switching and latency checks run on your machine, with no server in between.

Keys never uploaded

Credentials stay on disk, masked in the UI and never written to logs.

No admin rights

Installs and runs as a normal user. Nothing written to system folders, no background service.

Inspectable, reversible

Every switch backs up your original config first. Switching back restores it fully.

Built on

Native shell
barely noticeable once installed

Built with Tauri and Rust. No browser engine to carry. Starts fast, sits light in memory, closes clean.

7.5MB

Installer size

<1s

Cold start

2OS

Windows / macOS

Runs on Windows 10 and later and macOS 12 and later (one universal build for Intel and Apple silicon).

FAQ

Questions people actually ask

What does AiSwitch do?

It is a small Windows and macOS desktop app that switches AI coding clients between their official endpoints and third-party APIs in one click. It handles config files, environment variables and model mapping for you.

Which clients are supported?

Twelve today: ChatGPT (Codex), Claude Code, Claude Desktop, Antigravity, Grok Build, OpenCode, OpenClaw, MiniMax Code, ZCode, plus WorkBuddy, WorkBuddy China and WorkBuddy Global. More come with each release.

Will switching lose my existing config?

No. The app backs up the original config before every switch, and "Switch to official" restores it as it was. The backup is a plain file you can open yourself.

Are my API keys uploaded anywhere?

No. Reading and writing config, switching and latency checks all happen on your machine. There is no account system and no relay through our servers.

Is it free? Do I need an account?

Free, no account. On Windows it needs no admin rights — install it and open it.

How is this different from editing config.toml by hand?