One-command setup of ApiFlux for your local AI coding tools.
npx apiflux-cli initPaste your ApiFlux API key when prompted, pick the tools to configure, done. The CLI writes the correct base URL and key into each tool's config, verifies the key with a real request, and tells you exactly what it changed.
| Tool | What gets configured |
|---|---|
| Claude Code | ANTHROPIC_BASE_URL / ANTHROPIC_AUTH_TOKEN in ~/.claude/settings.json |
| Codex CLI | model_providers.apiflux in ~/.codex/config.toml (key stays in the APIFLUX_API_KEY env var, never in the file); with --model, also sets it as the default provider/model |
| OpenCode | provider.apiflux in ~/.config/opencode/opencode.json with every model your key can use; key goes to OpenCode's own credential store (~/.local/share/opencode/auth.json, 0600), never into the config file; with --model, also sets it as the default model |
| Pi | providers.apiflux in ~/.pi/agent/models.json with every model your key can use; key goes to Pi's own credential store (~/.pi/agent/auth.json, 0600), never into the config file; with --model, also persists it via defaultProvider/defaultModel in settings.json |
| Hermes | custom_providers entry in ~/.hermes/config.yaml (comment-preserving YAML edit) with every model your key can use; key goes to ~/.hermes/.env as APIFLUX_API_KEY (0600), never into config.yaml; with --model, also sets model.provider/model.default |
| OpenClaw | models.providers.apiflux in ~/.openclaw/openclaw.json with every model your key can use; with --model, also sets agents.defaults.model.primary (existing fallbacks kept). Note: like OpenClaw's own config writes, rewriting strips JSON5 comments; $include configs get a manual snippet instead |
| Anything OpenAI-compatible | Prints export OPENAI_BASE_URL / OPENAI_API_KEY (and OPENAI_MODEL if chosen) lines for your shell or .env |
Any model your key can use works in any tool — the ApiFlux gateway converts between the Anthropic and OpenAI protocols. Picking a non-Claude model for Claude Code also pins its small/fast background model so every request stays on your chosen model.
npx apiflux-cli init # interactive: hidden key prompt + tool picker
npx apiflux-cli init --key sk-xxx --yes # non-interactive (as copied from the ApiFlux console)
npx apiflux-cli init --key - # read the key from stdin| Flag | Meaning |
|---|---|
--key <key|-> |
API key; - reads from stdin; omit for a hidden prompt |
--base-url <url> |
Override the API origin (self-hosted / testing) |
--tool <id> |
Configure only this tool (repeatable): claude-code, codex, opencode, pi, hermes, openclaw, export |
--model <id> |
Default model for the configured tools; omit in a terminal to pick from a list, omit in scripts to keep each tool's default |
--yes |
Skip confirmations, overwrite conflicting config |
--skip-verify |
Don't send the verification request |
- Existing config files are backed up (
*.bak.<timestamp>) before any write; merges preserve your other settings; re-running is idempotent. - If a different base URL or token is already configured, the CLI asks before replacing it.
- Your key is never written to logs and is masked (
sk-***) in every error message. - If you pass
--keyinline, the CLI reminds you to clear it from shell history; prefer the interactive prompt or--key -. - The CLI never touches your ApiFlux account — it only ever holds the API key you give it, by design (ADR-0001).
Node.js ≥ 18. macOS and Linux supported; Windows via WSL.
Issues and pull requests are welcome — see CONTRIBUTING.md.