Frequently asked questions

yoink keeps every AI coding login and provider in one place. It switches Claude Code accounts, switches ChatGPT (Codex), Kimi Code, Gemini and GitHub Copilot logins inside their own tools, and writes API keys from presets or any compatible URL into up to 13 coding harnesses, depending on the endpoints the provider speaks. See Getting started.

No. Yoink for AI coding (yoink.codes, yoink-cli on npm, github.com/MajidRaimi/yoink) switches AI coding logins and providers. The Mac drag-and-drop shelf app and the PyPI media downloader are unrelated projects that happen to share the name. Install this one with npm install -g yoink-cli or the install script.

Run yoink add once per account to save each login, then move between them with yoink use personal (or the shortcut yoink work). Each switch stores the current login back into its profile before writing the target account's credential and identity, so you stay signed in. Restart Claude Code afterward. The Claude Code accounts guide has the details.

On macOS, in the Keychain entry Claude Code-credentials. On Linux and Windows, in the plaintext file ~/.claude/.credentials.json. The account identity lives separately, in the oauthAccount block of ~/.claude.json, and a yoink profile stores both halves together. See How it works.

Yes, from the CLI. Save each login with yoink add or yoink save <name> --tool <tool>, then switch with yoink use <name>. Each tool switches inside its own login home only, and yoink never copies a subscription token into another tool. See Subscriptions.

No. The menu bar app switches Claude Code accounts, saves the current Claude login and manages API-key providers. ChatGPT (Codex), Kimi Code, Gemini and GitHub Copilot logins switch from the CLI only today, with yoink use <name>. See Desktop app.

No. Every switch is one you start yourself, from a yoink use command, the interactive menu or the menu bar app. yoink has no view of your remaining quota and never picks a different account for you when a limit arrives. The switcher comparison lists tools that do.

Add the key once with yoink add, pick a preset or a custom URL, choose models, and tick the harnesses. Later, yoink connect <name> --to pi,opencode,codex adds more. yoink writes each harness's own config file and re-syncs them all when the key or models change. See Providers.

Thirteen: pi, omp, opencode, codex, Claude Code, Qwen Code, Kilo Code, Droid, Crush, Goose, Zed, Continue and Claude Desktop. Goose, Zed and Claude Desktop are experimental, so check the notice yoink prints after connecting them. The harness table lists each config path.

Yes. Both presets have an Anthropic Messages endpoint, which is what Claude Code needs. Connecting one writes seven managed env keys into ~/.claude/settings.json, and switching back to a Claude account strips them again. The ollama preset cannot be used there. See Claude Code is exclusive.

codex reaches custom providers only through the OpenAI Responses API. The openrouter preset offers OpenAI Chat and Anthropic Messages, not Responses, so codex shows "needs an OpenAI Responses endpoint". The openai preset is the one preset codex accepts. See codex is responses-only.

No. yoink edits the harness configs and then exits, leaving nothing running between your agent and the provider. Because nothing translates requests, a provider can only serve harnesses that already speak one of its protocols. Claude Code Router routes requests through a local gateway, and CC Switch can do the same in its optional Routing mode. See yoink vs Claude Code Router and yoink vs CC Switch.

Login switching swaps the Kimi Code subscription files in ~/.kimi-code/credentials/ and only affects Kimi Code itself; the token never leaves that tool. The kimi-code preset is an API key, written into other harnesses through https://api.kimi.com/coding/v1 (OpenAI Chat) or https://api.kimi.com/coding (Anthropic Messages). Compare Subscriptions and the preset table.

Only to the provider the key belongs to, when yoink probes that endpoint or asks it for a model list. The model limits come from the public models.dev catalog, which yoink downloads with no key attached and keeps for a day. Keys stay in ~/.config/yoink/profiles.json and the harness configs you connect. See Security.

Before its first edit to a harness config you already have, yoink saves the untouched file as <file>.yoink.bak, and later runs leave that copy alone. Every later write is atomic and owner-only, and yoink only touches the entries it owns. Claude Code's settings.json gets no backup, because only seven keys change. See Backups.

Not by accident. Before writing a key, yoink checks whether git tracks the config file it is about to change. The interactive flow names the tracked files and asks first; yoink add --external and yoink connect refuse unless you pass --allow-tracked. Backups inside a repo are added to .git/info/exclude. See Git-tracked configs.

Yes. Pass --token-stdin to yoink add --external, yoink edit or yoink probe and pipe the key in, which keeps it out of both your shell history and anything that lists running processes. The --json output of yoink list, yoink harnesses and yoink status never includes a token or key. See Keys stay off the command line.

Yes. yoink is open source under the MIT license. There are CLI builds for Intel and Apple Silicon Macs, for x64 and arm64 Linux with either glibc or musl, and for x64 and arm64 Windows. The menu bar app needs macOS 12 or later. See Supported platforms.

On macOS and Linux run curl -fsSL https://yoink.codes/install.sh | bash. On Windows run the PowerShell installer from Getting started. On any OS with Node, npm install -g yoink-cli works too. The Mac app is a DMG from the download page.

That prompt is macOS guarding the Keychain item that holds your Claude Code login. Answer it with Always Allow once and later switches run quietly; picking plain Allow means you see it again next time. The menu bar app triggers the same prompt. On Linux and Windows the login is a file, so nothing asks. See How it works.

A running tool can write its old token back on its next refresh. Quit Claude Code before switching and start it again afterward. For Codex, Kimi Code, Gemini and Copilot, yoink asks first when the tool is running, and a script must pass --force. See What a switch does.

On Linux, yoink does not support the Secret Service keyring yet, so it refuses instead of guessing when Codex keeps its login there. Set cli_auth_credentials_store = "file" in ~/.codex/config.toml, run codex login again so Codex writes auth.json, then retry the switch. See ChatGPT (Codex).

Two setups block it: GEMINI_FORCE_ENCRYPTED_FILE_STORAGE=true, or a login kept in gemini-credentials.json. In both cases the credential is not a portable file yoink can copy out and put back, so it refuses. Unset the variable and log in again with /auth, so Gemini keeps the login in oauth_creds.json, which yoink can switch. See Gemini.

When a crushrc file sits next to crush.json, it may override the provider yoink just wrote, so yoink prints a notice instead of failing. If Crush reads that file instead of crush.json and does not pick up the new provider, add the provider to crushrc too. See the Crush notes.