ai-provider-switcher

Point Claude Code, Codex CLI, OpenCode and Continue at a different LLM provider with one command — and find out where they currently point with another.

npx @apimaster/ai-provider-switcher status
npx @apimaster/ai-provider-switcher use apimaster

Zero dependencies, Node 18.17+.

The problem

If you use more than one coding agent, your provider config is spread across four files in four formats:

~/.claude/settings.json          env.ANTHROPIC_BASE_URL   (no /v1 — this one trips everybody up)
~/.codex/config.toml             [model_providers.x] base_url
~/.config/opencode/opencode.json provider.x.options.baseURL
~/.continue/config.yaml          models[].apiBase

Switching between a work gateway, a personal key and a local model means editing all four by hand, and when something breaks you cannot tell which tool is talking to which endpoint.

Commands

status — where is everything pointed

  TOOL            POINTS AT
  claude-code     ✔ https://apimaster.ai
                  ~/.claude/settings.json
  codex           ✔ https://apimaster.ai/v1
                  ~/.codex/config.toml
  opencode        ! http://localhost:11434    base_url has no /v1
                  ~/.config/opencode/opencode.json
  continue        not configured

use <profile> — switch everything at once

ai-switch use apimaster                    # every supported tool
ai-switch use anthropic --tools claude-code
ai-switch use ollama --tools codex,continue

Existing config files are copied into ~/.ai-switch/backups/ before anything is written, and ai-switch restore puts the last one back.

Tools that cannot serve a profile are skipped with a reason rather than half-configured: Claude Code speaks the Anthropic protocol, so a provider with no Anthropic-compatible endpoint is reported as skipped, not silently broken.

env <profile> — for a shell session only

eval "$(ai-switch env apimaster)"

add — your own gateway

ai-switch add work \
  --base-url https://gateway.internal/v1 \
  --anthropic-base-url https://gateway.internal \
  --key-env WORK_LLM_KEY \
  --model gpt-4o

Built-in profiles

| Profile | Base URL | Key from | | --- | --- | --- | | openai | https://api.openai.com/v1 | $OPENAI_API_KEY | | anthropic | https://api.anthropic.com/v1 | $ANTHROPIC_API_KEY | | apimaster | https://apimaster.ai/v1 | $APIMASTER_API_KEY | | openrouter | https://openrouter.ai/api/v1 | $OPENROUTER_API_KEY | | deepseek | https://api.deepseek.com/v1 | $DEEPSEEK_API_KEY | | ollama | http://localhost:11434/v1 | $OLLAMA_API_KEY |

Add your own with add; a profile you define shadows a built-in one with the same name.

Keys are never stored here

A profile records the name of an environment variable, not a key. When you run use, the key is read from your environment at that moment and written only into the tool's own config file, which is where the tool expects it. ~/.ai-switch/profiles.json stays free of secrets, so you can commit it to a dotfiles repo.

Keys that arrive wrapped in quotes or with a trailing newline — the single most common cause of a mysterious 401 — are trimmed, and the tool tells you it did.

Why ~/.claude/settings.json is different

Claude Code speaks Anthropic's Messages protocol, and its base URL is the site root:

{ "env": { "ANTHROPIC_BASE_URL": "https://apimaster.ai" } }

Adding /v1 there produces a 404 that looks like an auth problem. status flags it when it sees it.

Development

npm test
node bin/ai-switch.js status

License

MIT