claude-flow

Bridge any AI model to Claude Code CLI — route through OpenRouter, DeepSeek, OpenAI, Gemini, or any Anthropic-compatible endpoint.

Available in: English | Українська | Русский

The Problem

Claude Code CLI is hardcoded for Anthropic's API. It sends requests in Anthropic's Messages API format, expects responses in that exact format, and authenticates using Anthropic's method. If you want to use any other model — DeepSeek, GPT-4, Gemini, GLM, Llama, Qwen — you hit three walls:

1. Response Format Incompatibility

Every AI provider has its own API format. DeepSeek, OpenAI, and Gemini use the OpenAI-compatible format. Chinese models (GLM, Qwen) have their own APIs. Open-source models (Llama, Mistral) each have unique response structures.

Claude Code expects Anthropic's specific streaming events (content_block_delta), content block types, tool_use structures, stop_reason values, and token counting fields. Send it a raw DeepSeek or OpenAI response — it crashes.

The fix: Providers like OpenRouter and DeepSeek offer Anthropic-compatible proxy endpoints — they accept requests in Anthropic format, route them to any model, and translate the response back into Anthropic format. claude-flow knows exactly which endpoints to use for each provider.

2. Authentication Trap

Claude Code's auth has an undocumented quirk: ANTHROPIC_API_KEY must be set to an empty string ("") for proxy providers — not absent, not unset. Specifically an empty string.

If ANTHROPIC_API_KEY is absent → Claude Code errors out on startup
If ANTHROPIC_API_KEY is non-empty → Claude Code uses it and silently ignores your proxy token
If ANTHROPIC_API_KEY is "" → Claude Code falls back to ANTHROPIC_AUTH_TOKEN ✓

Different providers also use different auth methods — Bearer token (OpenRouter) vs. API key header (DeepSeek, OpenAI). Getting any of this wrong means silent failures.

3. Configuration Complexity

6 environment variables, zero official documentation, different combinations for each provider, and non-obvious behaviors. One wrong setting = Claude Code either crashes or silently talks to the wrong endpoint.

claude-flow solves all three. One command configures everything correctly for any provider.

Quick Start

# Install from GitHub
npm install -g github:Lexus2016/claude-flow

# Setup your provider (interactive)
claude-flow setup openrouter

# Run Claude Code
eval $(claude-flow env)
claude -p "Hello from OpenRouter!"

That's it. Now Claude Code uses your provider for every invocation.

Installation

Important: The claude-flow name on npm belongs to an unrelated project. Always install from GitHub to get the correct package.

Option 1: Install from GitHub (recommended)

npm install -g github:Lexus2016/claude-flow

Option 2: Clone and use locally

git clone https://github.com/Lexus2016/claude-flow.git
cd claude-flow
npm link
claude-flow setup openrouter

Requirements

Node.js >= 18.0
Claude Code CLI (from Anthropic) — install here
An API key for your chosen provider (see Supported Providers)

Step-by-Step Setup

1. Install claude-flow

npm install -g github:Lexus2016/claude-flow

2. Run the interactive setup

claude-flow setup openrouter

You'll be prompted for:

Your API key for the provider (e.g., sk-or-v1-...)
Model preferences for each tier (haiku, sonnet, opus) — or press Enter for defaults

Example session:

  claude-flow — Setup OpenRouter

  Get your API key: https://openrouter.ai/keys

  API Key: sk-or-v1-abc123def456...

  Model configuration (press Enter for defaults)

  Claude Code uses 3 model tiers: haiku (fast), sonnet (balanced), opus (powerful)

  Popular models for OpenRouter:
    anthropic/claude-sonnet-4-6             sonnet
    anthropic/claude-opus-4-6               opus
    anthropic/claude-haiku-4-5-20251001              haiku
    google/gemini-2.5-pro                 opus
    deepseek/deepseek-chat                sonnet

  Haiku model  [anthropic/claude-haiku-4-5-20251001]:
  Sonnet model [anthropic/claude-sonnet-4-6]:
  Opus model   [anthropic/claude-opus-4-6]:

  ✓ Configuration saved to /Users/you/.claude-flow/config.json
  ✓ Active provider: OpenRouter

  Environment variables (what Claude Code will use):

    ANTHROPIC_BASE_URL                 https://openrouter.ai/api
    ANTHROPIC_AUTH_TOKEN              sk-or-v1-abc...
    ANTHROPIC_API_KEY                 ""  ← intentionally empty
    ANTHROPIC_DEFAULT_HAIKU_MODEL     anthropic/claude-haiku-4-5-20251001
    ANTHROPIC_DEFAULT_SONNET_MODEL    anthropic/claude-sonnet-4-6
    ANTHROPIC_DEFAULT_OPUS_MODEL      anthropic/claude-opus-4-6

  Quick start:

    # Option 1: Run directly
    claude-flow run -- claude -p 'Hello from OpenRouter!'

    # Option 2: Add to your shell profile
    echo 'eval "$(claude-flow env)"' >> ~/.bashrc

    # Option 3: One-time eval
    eval $(claude-flow env)

3. Use Claude Code with your provider

Choose one of three methods:

Method 1: Run directly (recommended for one-off commands)

claude-flow run -- claude -p "What is the weather?"

Method 2: Eval to current shell (recommended for interactive work)

eval $(claude-flow env)
claude -p "Build me a React component"

Method 3: Add to shell profile (recommended for permanent setup)

# Add to ~/.bashrc, ~/.zshrc, or ~/.fish/config.fish
echo 'eval "$(claude-flow env)"' >> ~/.bashrc

# Then reload your shell
source ~/.bashrc

# Now Claude Code works everywhere
claude -p "Hello!"

Supported Providers

Quick Comparison

Provider	Model Selection	Cost	Setup	Popular For
OpenRouter	200+ models from all providers	Varies	Easy	One API for everything
DeepSeek	V3.2, R1 (reasoning)	Cheapest	Easy	Cost-effective reasoning
OpenAI	GPT-5.2, o3	Premium	Easy	Cutting-edge models
Gemini	Gemini 3.1 Pro / 3 Flash	Competitive	Easy	Fast inference
Custom	Any Anthropic-compatible endpoint	Varies	Manual	Self-hosted, private clouds

Provider Details

OpenRouter

Access 200+ models from a single API endpoint.

Perfect if you want flexibility — use Claude, GPT-4, DeepSeek, Gemini, Llama, Qwen, and more with a single key. Pay per use.

Get API key: https://openrouter.ai/keys

Setup:

claude-flow setup openrouter

Default models:

Haiku: anthropic/claude-haiku-4-5-20251001-5-20251001
Sonnet: anthropic/claude-sonnet-4-6-6
Opus: anthropic/claude-opus-4-6-6

Popular models to try:

claude-flow models openrouter

  Default tier mapping:
    haiku  → anthropic/claude-haiku-4-5-20251001-5-20251001
    sonnet → anthropic/claude-sonnet-4-6-6
    opus   → anthropic/claude-opus-4-6-6

  Popular models:

    Model ID                                 Name                     Tier
    ─────────────────────────────────────────────────────────────────────
    anthropic/claude-opus-4-6-6                Claude Opus 4.6          opus
    anthropic/claude-sonnet-4-6-6              Claude Sonnet 4.6        sonnet
    anthropic/claude-haiku-4-5-20251001-5-20251001      Claude Haiku 4.5         haiku
    google/gemini-3.1-pro-preview            Gemini 3.1 Pro           opus
    google/gemini-3-flash-preview            Gemini 3 Flash           sonnet
    openai/gpt-5.2                           GPT-5.2                  opus
    openai/gpt-5.2-mini                      GPT-5.2 Mini             sonnet
    deepseek/deepseek-v3.2                   DeepSeek V3.2            sonnet
    deepseek/deepseek-r1-0528                DeepSeek R1              opus
    minimax/minimax-m2.5                     MiniMax M2.5             opus
    moonshotai/kimi-k2.5                     Kimi K2.5                opus
    z-ai/glm-5                               GLM-5                    opus
    qwen/qwen3.5-plus                        Qwen 3.5                 sonnet
    meta-llama/llama-4-maverick              Llama 4 Maverick         sonnet
    z-ai/glm-4.5-air                         GLM-4.5 Air              haiku

Use different model per tier:

claude-flow setup openrouter

# When prompted:
# Haiku model: google/gemini-2.5-flash
# Sonnet model: deepseek/deepseek-v3.2
# Opus model: openai/gpt-5.2

DeepSeek

Ultra-fast and cheap. Great for reasoning tasks.

DeepSeek V3.2 (fast, hybrid thinking) and R1 (deep reasoning) through their native Anthropic-compatible API. Single digit cent costs.

Get API key: https://platform.deepseek.com/api_keys

Setup:

claude-flow setup deepseek

Default models:

Haiku: deepseek-chat
Sonnet: deepseek-v3.2
Opus: deepseek-r1-0528

Why use DeepSeek?

Cheapest LLM pricing available
DeepSeek V3.2 supports hybrid thinking mode
DeepSeek R1 reasoning beats most models on benchmarks
Native Anthropic-compatible API (no proxy markup)

Use R1 for deep reasoning:

claude-flow setup deepseek
# When prompted for Opus, use: deepseek-r1-0528

OpenAI

Latest GPT models, including GPT-5.2 and o3.

GPT-5.2 (flagship), GPT-5.2 Thinking (reasoning) and o3 through OpenAI's official API.

Get API key: https://platform.openai.com/api-keys

Setup:

claude-flow setup openai

Default models:

Haiku: gpt-5.2-mini
Sonnet: gpt-5.2
Opus: gpt-5.2

Popular models:

claude-flow models openai

  Default tier mapping:
    haiku  → gpt-5.2-mini
    sonnet → gpt-5.2
    opus   → gpt-5.2

  Popular models:

    Model ID                                 Name                     Tier
    ─────────────────────────────────────────────────────────────────────
    gpt-5.2                                  GPT-5.2                  opus
    gpt-5.2-mini                             GPT-5.2 Mini             sonnet
    gpt-5.2-thinking                         GPT-5.2 Thinking         opus
    o3                                       o3                       opus
    gpt-4.1                                  GPT-4.1                  sonnet

Gemini

Google's latest models. Fast and capable.

Gemini 3.1 Pro (latest flagship), Gemini 3 Flash, and Gemini 2.5 through Google's API.

Get API key: https://aistudio.google.com/apikey

Setup:

claude-flow setup gemini

Default models:

Haiku: gemini-2.5-flash
Sonnet: gemini-3-flash-preview
Opus: gemini-3.1-pro-preview

Custom Provider

Any Anthropic-compatible endpoint (self-hosted, private clouds, etc.).

If you have your own Claude-compatible endpoint, or want to use a provider not listed above:

claude-flow setup custom

You'll be prompted for:

API base URL (e.g., https://your-proxy.local/v1)
Model IDs
API key

CLI Reference

Help

claude-flow help

Shows all commands and examples.

Setup

claude-flow setup <provider>

Interactive setup wizard. Guides you through getting an API key and choosing models.

Providers: openrouter, deepseek, openai, gemini, custom

Examples:

claude-flow setup openrouter
claude-flow setup deepseek
claude-flow setup openai

Env

Print shell export statements. Used for eval or adding to your shell profile.

# Print to stdout (copy-paste friendly)
claude-flow env

# Or use in a command
eval $(claude-flow env)

# Or for a specific provider
claude-flow env deepseek
eval $(claude-flow env deepseek)

Output example:

export ANTHROPIC_BASE_URL='https://openrouter.ai/api'
export ANTHROPIC_AUTH_TOKEN='sk-or-v1-...'
export ANTHROPIC_API_KEY=''
export ANTHROPIC_DEFAULT_HAIKU_MODEL='anthropic/claude-haiku-4-5-20251001'
export ANTHROPIC_DEFAULT_SONNET_MODEL='anthropic/claude-sonnet-4-6'
export ANTHROPIC_DEFAULT_OPUS_MODEL='anthropic/claude-opus-4-6'

Run

Run a command with provider environment variables set.

claude-flow run -- <command> [args...]

The -- separator is important — everything after it is your command.

Examples:

# Run Claude Code once with OpenRouter
claude-flow run -- claude -p "Hello from OpenRouter!"

# Run with a different provider
claude-flow run deepseek -- claude -p "Use DeepSeek for this"

# Run any command (not just Claude)
claude-flow run -- node my-script.js
claude-flow run -- python my-app.py

# Chain with other commands
claude-flow run -- bash -c 'echo $ANTHROPIC_BASE_URL && claude -p "Test"'

Status

Show current configuration and active provider.

claude-flow status

Output example:

  claude-flow status

  Config: /Users/you/.claude-flow/config.json
  Active: openrouter

  ● OpenRouter         Key: sk-or-v1-abc...def
    haiku=anthropic/claude-haiku-4-5-20251001  sonnet=anthropic/claude-sonnet-4-6  opus=anthropic/claude-opus-4-6
  ○ DeepSeek           Key: sk-...xyz
    haiku=deepseek-chat  sonnet=deepseek-chat  opus=deepseek-reasoner

Switch

Switch to a different provider (must be configured first).

claude-flow switch <provider>

Examples:

claude-flow switch deepseek
# Now eval $(claude-flow env) will use DeepSeek

Providers

List all available providers with details.

claude-flow providers

Output:

  Available Providers

  ● openrouter       200+ models from every major provider through one API
     Auth: Bearer | Base: https://openrouter.ai/api | Key env: OPENROUTER_API_KEY
  ○ deepseek         DeepSeek models via native Anthropic-compatible API
     Auth: API key | Base: https://api.deepseek.com/anthropic | Key env: DEEPSEEK_API_KEY
  ○ openai           OpenAI models (GPT-4.1, o3, etc.)
     Auth: API key | Base: https://api.openai.com/v1 | Key env: OPENAI_API_KEY
  ○ gemini           Gemini models via OpenAI-compatible endpoint
     Auth: API key | Base: https://generativelanguage.googleapis.com/v1beta | Key env: GEMINI_API_KEY
  ○ custom           Any Anthropic-compatible API endpoint
     Auth: Bearer | Base: (custom) | Key env: CUSTOM_API_KEY

  ● = active provider

Models

Browse available models for a provider.

claude-flow models [provider]

If provider is omitted, shows models for the active provider.

Example:

claude-flow models openrouter
claude-flow models deepseek

Version

Show installed version.

claude-flow --version
# or
claude-flow -v

How It Works

Architecture

Without claude-flow:
  Claude Code CLI → Anthropic API → Claude models only
                                     (no other models possible)

With claude-flow:
  ┌─────────────┐    Anthropic    ┌──────────────────────┐    Native    ┌──────────────┐
  │ Claude Code │ ── Messages ──→ │ Provider's           │ ── API ───→ │ Any Model    │
  │ CLI         │    API format   │ Anthropic-compatible │    call      │              │
  │             │ ←─ Anthropic ── │ endpoint             │ ←─ Native ─ │ GPT, GLM,    │
  │             │    format       │                      │    format    │ Llama, Qwen, │
  └─────────────┘                 │ TRANSLATES responses │              │ DeepSeek,    │
                                  │ between formats      │              │ Gemini, ...  │
                                  └──────────────────────┘              └──────────────┘
                                            ▲
                                   claude-flow configures
                                   this connection correctly

The key insight: Claude Code doesn't need to talk to models directly. It talks to an Anthropic-compatible proxy endpoint that handles the format translation. OpenRouter translates 200+ models into Anthropic format. DeepSeek's /anthropic endpoint does the same. claude-flow knows which endpoint to use for each provider and configures Claude Code to connect there.

What Providers Do Behind the Scenes

When you use z-ai/glm-5 through OpenRouter:

Claude Code sends an Anthropic Messages API request to https://openrouter.ai/api/v1/messages
OpenRouter receives the request, translates it to GLM-5's native format
GLM-5 processes and returns a response in its own format
OpenRouter translates the response back into Anthropic Messages API format
Claude Code receives a response it understands — streaming events, content blocks, tool_use, everything

Without this translation, Claude Code gets a response in the wrong format and crashes.

The Environment Variables

Claude Code reads these env vars to determine which API to use:

Env Var	Purpose	Example
`ANTHROPIC_BASE_URL`	API endpoint (where requests go)	`https://openrouter.ai/api`
`ANTHROPIC_AUTH_TOKEN`	Bearer token (proxy providers)	`sk-or-v1-...`
`ANTHROPIC_API_KEY`	API key (direct providers) or `""`	`sk-...` or `""`
`ANTHROPIC_DEFAULT_HAIKU_MODEL`	Fast tier model	`anthropic/claude-haiku-4-5-20251001`
`ANTHROPIC_DEFAULT_SONNET_MODEL`	Balanced tier model	`anthropic/claude-sonnet-4-6`
`ANTHROPIC_DEFAULT_OPUS_MODEL`	Powerful tier model	`anthropic/claude-opus-4-6`

The Empty Key Trick

ANTHROPIC_API_KEY must be set to an empty string ("") for proxy providers, not absent or unset.

This is the most common failure point when configuring Claude Code with alternative providers.

Why?

If ANTHROPIC_API_KEY is absent/unset: Claude Code errors trying to read it
If ANTHROPIC_API_KEY is non-empty: Claude Code uses it and ignores ANTHROPIC_AUTH_TOKEN
If ANTHROPIC_API_KEY is empty string (""): Claude Code falls back to ANTHROPIC_AUTH_TOKEN

Example: OpenRouter (uses auth_token)

# ✓ CORRECT — Claude Code uses AUTH_TOKEN for OpenRouter
export ANTHROPIC_BASE_URL='https://openrouter.ai/api'
export ANTHROPIC_AUTH_TOKEN='sk-or-v1-...'
export ANTHROPIC_API_KEY=''  # Empty string, not absent!
export ANTHROPIC_DEFAULT_SONNET_MODEL='anthropic/claude-sonnet-4-6'

# ✗ WRONG — Missing API_KEY, Claude Code errors
export ANTHROPIC_BASE_URL='https://openrouter.ai/api'
export ANTHROPIC_AUTH_TOKEN='sk-or-v1-...'
# (no ANTHROPIC_API_KEY set at all)

# ✗ WRONG — Non-empty API_KEY, Claude Code ignores AUTH_TOKEN
export ANTHROPIC_BASE_URL='https://openrouter.ai/api'
export ANTHROPIC_AUTH_TOKEN='sk-or-v1-...'
export ANTHROPIC_API_KEY='some-random-value'

Example: DeepSeek (uses api_key directly)

# ✓ CORRECT — Claude Code uses API_KEY for DeepSeek
export ANTHROPIC_BASE_URL='https://api.deepseek.com/anthropic'
export ANTHROPIC_API_KEY='sk-...'  # The actual key
export ANTHROPIC_DEFAULT_SONNET_MODEL='deepseek-chat'

How claude-flow Gets It Right

The buildEnv() function in lib/env.js:

Selects the correct endpoint that handles format translation for each provider
For proxy providers (OpenRouter): Sets ANTHROPIC_API_KEY = "" and ANTHROPIC_AUTH_TOKEN = apiKey
For direct providers (DeepSeek, OpenAI): Sets ANTHROPIC_API_KEY = apiKey (their endpoints accept Anthropic format natively)
Maps models to tiers — sets ANTHROPIC_DEFAULT_HAIKU/SONNET/OPUS_MODEL
Returns a complete env object ready to pass to Claude Code — no guesswork, no gotchas

Configuration

Configuration is stored in ~/.claude-flow/config.json (600 permissions — owner read/write only).

File Format

{
  "activeProvider": "openrouter",
  "providers": {
    "openrouter": {
      "apiKey": "sk-or-v1-...",
      "models": {
        "haiku": "anthropic/claude-haiku-4-5-20251001",
        "sonnet": "anthropic/claude-sonnet-4-6",
        "opus": "anthropic/claude-opus-4-6"
      }
    },
    "deepseek": {
      "apiKey": "sk-...",
      "models": {
        "haiku": "deepseek-chat",
        "sonnet": "deepseek-chat",
        "opus": "deepseek-reasoner"
      }
    }
  }
}

Manual Editing

You can edit this file directly. Just remember:

Keep it secure (contains API keys)
Use valid JSON
Restart your shell for changes to take effect

Environment Variable Override

You can also set API keys via environment variables without using the config file:

export OPENROUTER_API_KEY='sk-or-v1-...'
export DEEPSEEK_API_KEY='sk-...'

# Then use claude-flow with those keys
claude-flow run -- claude -p "Hello"

JavaScript API

Use claude-flow as a library in your Node.js projects (e.g., Claude Code Studio, build tools, CI/CD).

Installation

npm install github:Lexus2016/claude-flow

API Overview

const {
  buildEnv,
  toShellExports,
  mergeEnv,
  getProvider,
  listProviders,
  PROVIDERS,
  config,
} = require('claude-flow');

`buildEnv(provider, opts)`

Build environment variables for Claude Code.

Arguments:

provider (string): Provider name (openrouter, deepseek, openai, gemini, custom)
opts (object):
- apiKey (string, required): Provider API key
- haiku (string): Model for haiku tier (overrides default)
- sonnet (string): Model for sonnet tier (overrides default)
- opus (string): Model for opus tier (overrides default)
- model (string): Use same model for all tiers (shorthand)
- baseUrl (string): Override API base URL (custom providers only)

Returns: Object with env vars

Example:

const { buildEnv } = require('claude-flow');

// Use default models
const env = buildEnv('openrouter', {
  apiKey: 'sk-or-v1-abc123...'
});

console.log(env);
// {
//   ANTHROPIC_BASE_URL: 'https://openrouter.ai/api',
//   ANTHROPIC_AUTH_TOKEN: 'sk-or-v1-abc123...',
//   ANTHROPIC_API_KEY: '',
//   ANTHROPIC_DEFAULT_HAIKU_MODEL: 'anthropic/claude-haiku-4-5-20251001',
//   ANTHROPIC_DEFAULT_SONNET_MODEL: 'anthropic/claude-sonnet-4-6',
//   ANTHROPIC_DEFAULT_OPUS_MODEL: 'anthropic/claude-opus-4-6'
// }

Custom models:

const env = buildEnv('openrouter', {
  apiKey: 'sk-or-v1-...',
  haiku: 'google/gemini-2.5-flash',
  sonnet: 'deepseek/deepseek-chat',
  opus: 'openai/gpt-4.1'
});

Single model for all tiers:

const env = buildEnv('openrouter', {
  apiKey: 'sk-or-v1-...',
  model: 'deepseek/deepseek-chat'  // All tiers use this
});

Custom endpoint:

const env = buildEnv('custom', {
  apiKey: 'sk-custom-key',
  baseUrl: 'https://your-proxy.local/v1',
  haiku: 'your-model-haiku',
  sonnet: 'your-model-sonnet',
  opus: 'your-model-opus'
});

`mergeEnv(provider, opts)`

Build provider env vars and merge them into process.env. Returns a new object (doesn't mutate process.env).

Useful for spawn() with custom env.

Example:

const { spawn } = require('child_process');
const { mergeEnv } = require('claude-flow');

const env = mergeEnv('openrouter', {
  apiKey: 'sk-or-v1-...'
});

// Run Claude Code with OpenRouter
const child = spawn('claude', ['-p', 'Hello!'], {
  env,  // Use the merged env
  stdio: 'inherit'
});

`toShellExports(env)`

Format env vars as shell export statements.

Arguments:

env (object): Env object from buildEnv()

Returns: String with shell exports (one per line)

Example:

const { buildEnv, toShellExports } = require('claude-flow');

const env = buildEnv('openrouter', { apiKey: 'sk-or-v1-...' });
const shellScript = toShellExports(env);

console.log(shellScript);
// export ANTHROPIC_BASE_URL='https://openrouter.ai/api'
// export ANTHROPIC_AUTH_TOKEN='sk-or-v1-...'
// export ANTHROPIC_API_KEY=''
// export ANTHROPIC_DEFAULT_HAIKU_MODEL='anthropic/claude-haiku-4-5-20251001'
// export ANTHROPIC_DEFAULT_SONNET_MODEL='anthropic/claude-sonnet-4-6'
// export ANTHROPIC_DEFAULT_OPUS_MODEL='anthropic/claude-opus-4-6'

`getProvider(name)`

Get provider configuration by name.

Arguments:

name (string): Provider name or alias (openrouter, or, deepseek, ds, openai, gpt, gemini, custom)

Returns: Provider config object, or null if unknown

Example:

const { getProvider } = require('claude-flow');

const provider = getProvider('openrouter');
console.log(provider);
// {
//   name: 'OpenRouter',
//   description: '200+ models from every major provider...',
//   envKey: 'OPENROUTER_API_KEY',
//   baseUrl: 'https://openrouter.ai/api',
//   auth: 'auth_token',
//   models: { haiku: '...', sonnet: '...', opus: '...' },
//   docsUrl: 'https://openrouter.ai/keys',
//   popularModels: [...]
// }

`listProviders()`

List all provider names (excludes aliases).

Returns: Array of provider names

Example:

const { listProviders } = require('claude-flow');

console.log(listProviders());
// ['openrouter', 'deepseek', 'openai', 'gemini', 'custom']

`PROVIDERS`

Direct access to the provider configurations object.

Example:

const { PROVIDERS } = require('claude-flow');

console.log(PROVIDERS.openrouter.docsUrl);
// 'https://openrouter.ai/keys'

console.log(PROVIDERS.openrouter.popularModels);
// [{ id: 'anthropic/claude-sonnet-4-6', name: '...', tier: 'sonnet' }, ...]

`config`

Config file management (for integration with other tools).

Methods:

config.load() — Load entire config from disk
config.save(cfg) — Save config to disk
config.getActiveProvider() — Get active provider name
config.setActiveProvider(name) — Set active provider
config.getProviderConfig(name) — Get settings for a provider
config.setProviderConfig(name, settings) — Save settings for a provider
config.getConfigPath() — Get config file path

Example:

const { config } = require('claude-flow');

const activeProvider = config.getActiveProvider();
const settings = config.getProviderConfig(activeProvider);

console.log(settings.apiKey);    // The saved API key
console.log(settings.models);    // { haiku: '...', sonnet: '...', opus: '...' }

Integration Examples

Claude Code Studio

Claude Code Studio uses claude-flow to let you choose LLM providers in the UI.

const { buildEnv, PROVIDERS } = require('claude-flow');

// In your app, let user pick a provider
const selectedProvider = 'openrouter';
const apiKey = user.apiKeys[selectedProvider];

// Build env vars
const env = buildEnv(selectedProvider, { apiKey });

// Pass to Claude Code subprocess
spawn('claude', args, { env: mergeEnv(selectedProvider, { apiKey }) });

Docker / CI-CD

Use claude-flow in your Docker container to route to any LLM:

FROM node:18-slim

RUN npm install -g github:Lexus2016/claude-flow

WORKDIR /app
COPY . .

# Run Claude Code with provider from env
CMD ["sh", "-c", "eval $(claude-flow env) && claude -p 'Your prompt'"]

Build and run:

docker build -t my-app .

docker run --env OPENROUTER_API_KEY='sk-or-v1-...' my-app

Or use docker-compose.yml:

version: '3'
services:
  app:
    build: .
    environment:
      OPENROUTER_API_KEY: ${OPENROUTER_API_KEY}
      ANTHROPIC_DEFAULT_SONNET_MODEL: deepseek/deepseek-chat

Shell Profile (Permanent)

Add to your ~/.bashrc, ~/.zshrc, or ~/.fish/config.fish:

# Activate Claude Code provider
eval "$(claude-flow env)"

Now Claude Code will use your configured provider for every invocation in new shells.

Reload your shell:

source ~/.bashrc  # or ~/.zshrc, etc.

Troubleshooting

"Unknown command"

Make sure claude-flow is installed and in your PATH:

which claude-flow
# If empty, install from GitHub:
npm install -g github:Lexus2016/claude-flow

"No active provider"

You haven't run setup yet:

claude-flow setup openrouter
# or another provider

"No API key for provider"

The provider is configured but the API key wasn't saved. Run setup again:

claude-flow setup openrouter

Or manually add the key to ~/.claude-flow/config.json.

Claude Code still uses Anthropic's API

Make sure you've:

Run claude-flow setup <provider>
Either:
- Run eval $(claude-flow env) in your current shell
- Or use claude-flow run -- claude ... to run Claude Code once
- Or add eval "$(claude-flow env)" to your shell profile and restart

Check your active provider:

claude-flow status

"Key doesn't start with expected prefix"

This is a warning, not an error. It means your API key doesn't match the expected format for that provider. Double-check it's the right key:

OpenRouter keys start with sk-or-
DeepSeek keys start with sk-
OpenAI keys start with sk-
Gemini keys start with AI

If you're sure it's correct, you can ignore the warning.

Custom provider not working

For custom endpoints, make sure:

The base URL includes /v1 or the full path to the API (e.g., https://your-proxy.local/v1)
The model IDs are valid for your endpoint
Your API key is correct

Run:

claude-flow status

And verify all settings.

"ANTHROPIC_API_KEY" related errors

If Claude Code errors with "invalid API key" or similar, it might be using the wrong auth method. Check what you configured:

claude-flow env

Look at the output:

If you see ANTHROPIC_AUTH_TOKEN = (not empty) and ANTHROPIC_API_KEY = '' (empty quotes) → You're using a proxy provider correctly ✓
If you see ANTHROPIC_API_KEY = (your key) and no ANTHROPIC_AUTH_TOKEN → You're using a direct provider ✓
If you see both with values → There's a conflict, reconfigure

Testing

Run the test suite:

npm test

Tests cover:

Provider configuration loading
Environment variable building for each provider
The critical "empty string" behavior for proxy providers
Shell export formatting
Custom model overrides
Error handling

All tests pass with zero external dependencies.

How to Contribute

Fork the repo
Create a branch for your feature
Write tests for new functionality
Submit a pull request

Adding a new provider

Add to PROVIDERS in lib/providers.js
Include name, description, envKey, baseUrl, auth type, default models, and docsUrl
Add test cases in test/test.js
Update this README with the provider section

License

MIT — See LICENSE file

Related Projects

Claude Code CLI — The CLI we're configuring
Claude Code Studio — IDE that uses claude-flow
OpenRouter — Access 200+ models
DeepSeek Platform — Ultra-cheap reasoning models

Support

GitHub Issues: Report bugs or request features at https://github.com/Lexus2016/claude-flow/issues
Discussions: https://github.com/Lexus2016/claude-flow/discussions
Author: CDZV — Code Zero Digital Visual Trading

Made with love for developers who want to use any LLM with Claude Code.

Name		Name	Last commit message	Last commit date
Latest commit History 10 Commits
.github		.github
bin		bin
lib		lib
test		test
.gitignore		.gitignore
LICENSE		LICENSE
README.md		README.md
README_RU.md		README_RU.md
README_UA.md		README_UA.md
package.json		package.json

Uh oh!

Folders and files

Latest commit

History

Repository files navigation

claude-flow

The Problem

1. Response Format Incompatibility

2. Authentication Trap

3. Configuration Complexity

Quick Start

Installation

Option 1: Install from GitHub (recommended)

Option 2: Clone and use locally

Requirements

Step-by-Step Setup

1. Install claude-flow

2. Run the interactive setup

3. Use Claude Code with your provider

Method 1: Run directly (recommended for one-off commands)

Method 2: Eval to current shell (recommended for interactive work)

Method 3: Add to shell profile (recommended for permanent setup)

Supported Providers

Quick Comparison

Provider Details

OpenRouter

DeepSeek

OpenAI

Gemini

Custom Provider

CLI Reference

Help

Setup

Env

Run

Status

Switch

Providers

Models

Version

How It Works

Architecture

What Providers Do Behind the Scenes

The Environment Variables

The Empty Key Trick

Why?

Example: OpenRouter (uses auth_token)

Example: DeepSeek (uses api_key directly)

How claude-flow Gets It Right

Configuration

File Format

Manual Editing

Environment Variable Override

JavaScript API

Installation

API Overview

buildEnv(provider, opts)

mergeEnv(provider, opts)

toShellExports(env)

getProvider(name)

listProviders()

PROVIDERS

config

Integration Examples

Claude Code Studio

Docker / CI-CD

Shell Profile (Permanent)

Troubleshooting

"Unknown command"

"No active provider"

"No API key for provider"

Claude Code still uses Anthropic's API

"Key doesn't start with expected prefix"

Custom provider not working

"ANTHROPIC_API_KEY" related errors

Testing

How to Contribute

Adding a new provider

License

Related Projects

`buildEnv(provider, opts)`

`mergeEnv(provider, opts)`

`toShellExports(env)`

`getProvider(name)`

`listProviders()`

`PROVIDERS`

`config`

Packages