AI Switch

Local AI API Key hub — change once, every tool works

Why · Quick Start · What it does · Docs · Backends · License

Why

If you juggle relay / multi-vendor API keys across OpenClaw, OpenCode, Codex CLI, Claude Code… you already know the pain:

Pain	Without AI Switch	With AI Switch
Rotate one key	Edit 5 config files by hand	Change once → push to installed tools
Bad key	Tool hangs / weird 401s	Health check → strip from backends
Wrong write	Silent overwrite	Sync preview · uninstalled = no write
Onboarding	“Where do I put the key?”	3-minute path: add → check → push → test

Positioning: a local Key hub (manage · probe · push) — not another chat client.

One key change. Every installed tool updates. Broken keys don’t take your stack down.

Quick Start

Easiest: download a build from Releases
(Windows / macOS Intel / Apple Silicon / Linux) → extract → run ai-switch / start.bat / start.sh
→ open http://127.0.0.1:8787

From source:

git clone https://github.com/kingkate2009-droid/ai-switch.git
cd ai-switch
pip install -r requirements.txt
python3 run.py
# → http://127.0.0.1:8787

Docker:

docker compose up -d
# → http://127.0.0.1:8787

Requires Python 3.9+ for source installs.

3-minute loop: docs/quickstart.md

Add vendor + key (or smart import)
Health check passes
Push to backends (preview first)
Send hi in OpenCode / Codex / Claude Code

What it does

Only three outcomes matter day-to-day:

One place for keys — vendors, tags, batch import, MetaAPI merge, dedupe
Know what’s alive — layered probes (Chat + Codex Responses), readable errors, adaptive interval
Safe sync — preview · last-sync summary · never write to uninstalled tools · active provider switch

Also included when you need them: check-in URLs, budgets, encrypted profile export, diagnostics pack, downstream routes, desktop packages.

Feature list (expand)

Multi-vendor + key table, search / tags / batch enable-disable-delete
26+ built-in providers + custom OpenAI-compatible endpoints
Health monitor (scheduled), auto-disable / failover (optional)
Sync to OpenClaw, OpenCode, Claude Code, Codex, Cline, Aider, Continue, …
Batch / backup / MetaAPI import with undo
Light/dark theme, EN / 简体 / 繁中
Drop-in backend adapters (contribution guide)

Docs

Doc	For
3-minute quickstart	First run
Codex won’t connect	Responses / env / models
Backends	What each engine supports
Adapter contribution	Add a backend

Issues / help: open an issue

Supported backends

Backend	Notes
OpenClaw	`openclaw.json`, auth profiles, models
OpenCode	`auth.json`, `opencode.jsonc`
Claude Code	`settings.json`, `~/.claude.json`
Codex CLI	`auth.json`, `config.toml` + Responses health
Cline / Aider / Continue.dev	IDE / CLI configs
Hermes Agent, QwenCode, Kimi Code, TRAE Work	See backends.md

Status on the Backends page: Not installed · Stopped (syncable) · Running — uninstalled tools are never written.

Providers (probe types)

Type	Examples
`openai_chat`	OpenAI, DeepSeek, OpenRouter, Groq, Moonshot, Qwen, …
`anthropic`	Anthropic-compatible
`gemini`	Google Gemini
`openai_responses`	Codex-oriented `/v1/responses` path

Desktop packages

Platform	Asset pattern on Releases
Windows x64	`ai-switch-<ver>-windows-amd64.zip`
macOS Intel	`ai-switch-<ver>-macos-amd64.tar.gz`
macOS Apple Silicon	`ai-switch-<ver>-macos-arm64.tar.gz`
Linux x64	`ai-switch-<ver>-linux-amd64.tar.gz`

# Local package
bash scripts/build_package.sh
# Multi-platform release via CI
./release.sh v2.0.4

Configuration & security

Item	Path
Data	`~/.ai-switch/data.json`
Usage	`~/.ai-switch/usage.json`
Port	`8787` (`AI_SWITCH_PORT`)

Keys live only under ~/.ai-switch/. Never commit that directory. Optional access token in Settings for LAN exposure.

Stack: Python + Flask · vanilla JS · JSON storage · pluggable adapters · Apache 2.0

License

Apache 2.0

ai-switch