CrewX 0.6.0 Release - API Provider Support

November 17, 2025 · 6 min read

Founder @ SowonLabs

CrewX 0.6.0 is here! This release significantly expands CrewX's AI provider ecosystem by adding API Provider support. Now you can use OpenAI, Anthropic, Ollama, and other API-based AI providers directly, in addition to CLI-based providers. OpenRouter is also supported via OpenAI-compatible API.

🎯 Key Features

1. BYOA (Bring Your Own API) - Use Your Own APIs

CrewX has always embraced the BYOA (Bring Your Own AI) philosophy. You could leverage your existing AI subscriptions through CLI providers (Claude Code, Gemini Code Assist, GitHub Copilot). Now, with the addition of API Providers, this philosophy has evolved further.

Existing BYOA (CLI Providers):

agents:
  - id: "my_agent"
    provider: "cli/claude"    # Using Claude Code CLI

New BYOA (API Providers):

agents:
  - id: "my_agent"
    provider: "api/anthropic"  # Using Anthropic API directly
    inline:
      apiKey: "{{env.ANTHROPIC_API_KEY}}"
      model: "claude-sonnet-4-5-20250929"

Now you can use the AI API subscriptions you already own directly in CrewX. No additional cost, with your own APIs.

2. API Provider Support

CrewX 0.6.0 supports the following API providers:

✅ Currently Supported (Tested)

api/openai - OpenAI & OpenRouter

Support for GPT-4, GPT-4 Turbo, GPT-3.5 models
Automatic OpenRouter detection (when baseURL contains openrouter.ai)
Direct use with OpenAI API key

# Using OpenAI directly
- id: "gpt_agent"
  provider: "api/openai"
  inline:
    apiKey: "{{env.OPENAI_API_KEY}}"
    model: "gpt-4-turbo-preview"

# Using OpenRouter
- id: "openrouter_agent"
  provider: "api/openai"
  inline:
    baseURL: "https://openrouter.ai/api/v1"
    apiKey: "{{env.OPENROUTER_API_KEY}}"
    model: "anthropic/claude-3.5-sonnet"

api/anthropic - Anthropic Claude API

Support for Claude 3.7 Sonnet, Claude 3.5 Sonnet, Claude 3 Opus, and Haiku
Direct use with Anthropic API key

- id: "claude_agent"
  provider: "api/anthropic"
  inline:
    apiKey: "{{env.ANTHROPIC_API_KEY}}"
    model: "claude-sonnet-4-5-20250929"

api/ollama - Ollama (Local Models)

Local open-source models like Llama, Mistral
Run locally without internet connection
Free to use

- id: "ollama_agent"
  provider: "api/ollama"
  inline:
    baseURL: "http://localhost:11434"
    model: "llama3.2"

🔮 Coming Soon

These providers will be added in future versions:

Google Gemini (api/google) - Support for Gemini Pro, Ultra
AWS Bedrock (api/bedrock) - Claude, Titan, Llama, and more
LiteLLM (api/litellm) - Unified proxy for 100+ AI providers

3. Tool Calling - AI Can Read and Write Files Directly

API Providers come with built-in Tool Calling capabilities. AI agents can now use the following tools:

Query Mode (Read-Only)

read_file: Read files
grep: Pattern search
ls: Directory listing

Execute Mode (Write-Enabled)

write_file: Write files
replace: Replace text
run_shell_command: Execute shell commands

How it works:

User requests "Read the README.md file for me"
AI calls the read_file tool to fetch file content
AI analyzes the content and responds

Query vs Execute Mode:

Query Mode: Only read-only tools available (safe)
Execute Mode: Can modify files and run shell commands (use with care)

# Query mode - read-only
crewx query "@my_agent Analyze the README.md file"

# Execute mode - can modify files
crewx execute "@my_agent Add an installation section to README.md"

4. Runtime Model Override

Specify a default model in agent configuration, but use a different model when needed:

agents:
  - id: "smart_agent"
    provider: "api/anthropic"
    inline:
      apiKey: "{{env.ANTHROPIC_API_KEY}}"
      model: "claude-3-5-sonnet-20241022"  # Default model

# Use default model
crewx q "@smart_agent Simple question"

# Override with Opus model (for complex tasks)
crewx q "@smart_agent:claude-3-opus-20240229 Design complex architecture"

# Override with Haiku model (for fast response)
crewx q "@smart_agent:claude-3-haiku-20240307 Quick question"

📋 CLI Provider vs API Provider Comparison

Feature	CLI Provider	API Provider
Setup	CLI tool installation required	API key only
Authentication	Managed by CLI tool	Provide API key directly
Models	Specified by CLI tool	Specified in `agents.yaml`
Tool Calling	Depends on CLI tool	Built-in CrewX tools
Cost	Requires CLI subscription	Based on API usage
Network	CLI tool → API	Direct API calls
Flexibility	Limited by CLI tool capabilities	Full API functionality

When to Use What?

Recommended for CLI Provider:

✅ Already using Claude Code, Gemini Code Assist, GitHub Copilot
✅ Need IDE integration and additional CLI tool features
✅ Want to delegate authentication management to CLI tool

Recommended for API Provider:

✅ Want to manage API keys directly
✅ Want to freely switch between models
✅ Use local models like Ollama
✅ Use LiteLLM for unified multi-provider management
✅ Want to leverage Tool Calling capabilities to the fullest

🚀 Quick Start Examples

Example 1: Code Review Agent with Anthropic API

agents:
  - id: "code_reviewer"
    provider: "api/anthropic"
    inline:
      apiKey: "{{env.ANTHROPIC_API_KEY}}"
      model: "claude-sonnet-4-5-20250929"
      prompt: |
        You are a professional code reviewer.
        Find bugs, performance issues, and security vulnerabilities
        and suggest improvements.

# Review file (Query mode - read-only)
crewx q "@code_reviewer Review src/app.ts"

# Perform fixes (Execute mode)
crewx x "@code_reviewer Fix security issues in src/app.ts"

Example 2: Multi-Agent with OpenAI + Ollama

agents:
  - id: "gpt_architect"
    provider: "api/openai"
    inline:
      apiKey: "{{env.OPENAI_API_KEY}}"
      model: "gpt-4-turbo-preview"
      prompt: |
        You are a system architect.

  - id: "local_coder"
    provider: "api/ollama"
    inline:
      baseURL: "http://localhost:11434"
      model: "llama3.2"
      prompt: |
        You are a code implementation expert.

# Architecture design with GPT-4
crewx q "@gpt_architect Design microservices architecture"

# Code implementation with local Llama (cost savings)
crewx x "@local_coder Implement API endpoints"

# Query both agents simultaneously
crewx q "@gpt_architect @local_coder Analyze pros and cons of this design"

Example 3: Unified Provider Management with LiteLLM

agents:
  - id: "litellm_agent"
    provider: "api/litellm"
    inline:
      baseURL: "http://localhost:4000"
      apiKey: "{{env.LITELLM_API_KEY}}"
      model: "gpt-4"  # Routed by LiteLLM

# Start LiteLLM server
litellm --config litellm_config.yaml

# Use in CrewX
crewx q "@litellm_agent Your question here"

📦 Additional Improvements

Environment Variable Substitution: Safe API key management using {{env.VAR}} syntax
Inline Configuration Priority: Override root-level settings per agent
Agent ID Extraction: Accurately parse API agents from YAML
Improved Parallel Processing: Normalize provider formats during multi-agent execution

🔧 How to Upgrade

# Update via NPM
npm update -g crewx

# Or reinstall
npm install -g crewx@latest

# Verify version
crewx --version  # Should output 0.6.0

Your existing crewx.yaml configuration will continue to work as before. To use API Providers, simply add agents with the provider: "api/..." format as shown in the examples above.

📚 Next Steps

Upgrade to CrewX 0.6.0 and try API Providers!

# 1. Set API keys
export ANTHROPIC_API_KEY="your-key"
export OPENAI_API_KEY="your-key"

# 2. Configure crewx.yaml
vim crewx.yaml

# 3. Run agents
crewx q "@my_agent Hello!"

Have questions? Leave them on GitHub Issues! 🙌

🎯 Key Features​

1. BYOA (Bring Your Own API) - Use Your Own APIs​

2. API Provider Support​

✅ Currently Supported (Tested)​

🔮 Coming Soon​

3. Tool Calling - AI Can Read and Write Files Directly​

Query Mode (Read-Only)​

Execute Mode (Write-Enabled)​

4. Runtime Model Override​

📋 CLI Provider vs API Provider Comparison​

When to Use What?​

🚀 Quick Start Examples​

Example 1: Code Review Agent with Anthropic API​

Example 2: Multi-Agent with OpenAI + Ollama​

Example 3: Unified Provider Management with LiteLLM​

📦 Additional Improvements​

🔧 How to Upgrade​

📚 Next Steps​