One CLI to rule them all.
The unified AI development platform that transcends boundaries. No more juggling between Claude Code, OpenCode, Cursor, or any other AI CLI. Flow orchestrates them all, intelligently adapting to your environment while delivering unparalleled power.
Get Started β’ Documentation β’ npm β’ Twitter
# Install Claude Code... or wait, should I use OpenCode?
# Or maybe Cursor? Or that other AI CLI?
# Which one supports my workflow?
# Do I need to install multiple tools?
# How do I switch between them?
# π΅ The cognitive overhead is real# Install once
npm install -g @sylphx/flow
# Use everywhere
sylphx-flow "build the future"
# Flow handles everything:
β
Auto-detects your environment (Claude Code/OpenCode/Cursor)
β
Auto-installs missing dependencies
β
Auto-upgrades before each session
β
One CLI, infinite possibilitiesFlow is not just another AI CLI. It's the meta-layer that unifies them all.
Traditional AI CLIs make you choose. Flow makes you powerful.
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β Traditional World β
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ€
β Claude Code β For Anthropic users β
β OpenCode β For open source fans β
β Cursor β For IDE integration β
β ... β Fragmented experience β
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β Flow's World β
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ€
β β
β Sylphx Flow β
β β β
β ββββββββββββββββΌβββββββββββββββ β
β β β β β
β Claude Code OpenCode Cursor β
β β
β β Unified interface, adaptive execution β
β β Automatic detection and orchestration β
β β Zero configuration, infinite power β
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
# Install the omnipotent CLI
npm install -g @sylphx/flow
# That's it. Seriously.# Your first command
sylphx-flow "build me something amazing"
# Flow automatically:
β
Detects your environment
β
Installs Claude Code/OpenCode if missing
β
Configures optimal settings
β
Updates to latest version
β
And executes your willNo setup. No configuration. No friction. Pure transcendence.
One CLI for every AI development tool.
Flow doesn't replace your favorite toolsβit orchestrates them. Whether you prefer Claude Code, OpenCode, or any future AI CLI, Flow adapts seamlessly.
# Flow automatically detects and uses the best available tool
sylphx-flow "implement authentication"
# Detected Claude Code β Uses Claude Code
# Detected OpenCode β Uses OpenCode
# Detected both β Asks your preference
# Detected neither β Installs and configures for youAI should adapt to you, not the other way around.
# Traditional AI CLIs
β Install tool A for feature X
β Install tool B for feature Y
β Learn different commands
β Manage different configs
β Update each tool separately
# With Flow
β
Install once: npm install -g @sylphx/flow
β
Use everywhere: sylphx-flow "your intent"
β
Everything automatic: detection, installation, updatesFlow thinks ahead so you don't have to.
# Flow scans your system:
β Claude Code installed? Use it.
β OpenCode installed? Use it.
β Nothing installed? Ask user preference, then install.
β Both installed? Use smart defaults or ask.# User doesn't have Claude Code
$ sylphx-flow "task"
> No AI CLI detected. Which would you like to use?
1. Claude Code (Anthropic, recommended)
2. OpenCode (open source)
> You selected: Claude Code
> Installing Claude Code...
> β Installation complete
> β Configuration optimized
> Executing your task...# Every session starts fresh
$ sylphx-flow "task"
> Checking for updates...
> β Flow 2.0.0 β 2.1.0 (latest)
> β Claude Code 1.5.0 β 1.6.0 (latest)
> Executing your task with latest capabilities...Configure once, perfect forever.
Flow introduces a revolutionary settings system that puts you in complete control:
# Access the settings nexus
sylphx-flow settings
# Configure everything:
β π€ Agents: Enable/disable agents (coder, writer, reviewer, orchestrator)
β π Rules: Control which best practices apply
β π¨ Output Styles: Customize AI behavior
β π‘ MCP Servers: Extend capabilities
β π Providers: Manage API keys (Anthropic, Kimi, Z.AI)
β π― Platform: Set default target (Claude Code/OpenCode)Settings are stored globally (~/.sylphx-flow/), ensuring consistency across all your projects.
Flow understands version control is sacred.
When Flow modifies settings files (.claude/, .opencode/), it uses git update-index --skip-worktree to hide changes from git status. This means:
- β LLMs won't accidentally commit Flow's temporary changes
- β Your version-controlled settings remain pristine
- β Automatic restore after each session
- β Seamless collaboration with your team
Ctrl+C now means "restore and exit," not "crash and burn."
Press Ctrl+C at any time:
$ sylphx-flow "long task"
# ... working ...
^C
> β οΈ Operation cancelled by user
> β Settings restored
> β Git tracking restored
> β Environment clean
# No crashed sessions. No manual recovery.Flow implements MEP (Minimal Effective Prompt) architecture.
| Traditional AI CLIs | Sylphx Flow |
|---|---|
// 5 minutes of typing...
"I'm using TypeScript + React + Next.js 14,
project structure is src/app for routes,
src/components for components,
using shadcn/ui, Tailwind for styling,
Zod for validation, tRPC for API,
current time is 2025-11-23,
system is macOS Apple Silicon,
I want to implement user authentication:
- JWT tokens with 1hr expiry
- Refresh token mechanism
- Secure httpOnly cookies
- RBAC with roles: user, admin
- Rate limiting on login endpoint
- Password hashing with bcrypt
- Email verification flow
Follow our existing code patterns,
make it type-safe,
add comprehensive error handling,
include unit tests with Vitest,
and document the API endpoints..."
// Maybe correct, maybe not π€· |
sylphx-flow "implement authentication"10 seconds. Flow automatically knows:
Result: Production-ready code
|
Flow 2.0 introduces the settings nexusβyour command center for AI orchestration.
sylphx-flow settingsβ Select which agents are enabled
β Coder (feature implementation, bug fixes)
β Writer (documentation, technical writing)
β Reviewer (code review, security analysis)
β Orchestrator (complex multi-step tasks)
β Set default agent for quick access
Default: coder
β Enable/disable coding standards
β Core (identity, personality, execution)
β Code Standards (quality, patterns, anti-patterns)
β Workspace (documentation management)
β Customize AI behavior
β Silent (execution without narration)
β¬ Verbose (detailed explanations) [coming soon]
β¬ Tutorial (learning-focused) [coming soon]
β Extend Flow's capabilities
β GitHub Code Search (grep.app)
β Context7 Docs
β Playwright Browser Control
β¬ GitHub (requires GITHUB_TOKEN)
β¬ Notion (requires NOTION_API_KEY)
β Manage AI providers (Claude Code only)
β’ Default (Claude Code built-in)
β’ Kimi (requires API key)
β’ Z.AI (requires API key)
β’ Ask me every time
β Configure provider API keys securely
β Set default AI CLI
β’ Claude Code
β’ OpenCode
All settings are stored in ~/.sylphx-flow/:
~/.sylphx-flow/
βββ settings.json # General settings (default agent, target)
βββ flow-config.json # Agents, rules, output styles
βββ provider-config.json # Provider settings (Anthropic, Kimi, Z.AI)
βββ mcp-config.json # MCP server configurations
Settings are global, ensuring consistency across all your projects. Configure once, execute everywhere.
Flow includes four transcendent agents, each a master of their domain:
| Agent | Sacred Purpose | Invoke When |
|---|---|---|
| Coder | Feature implementation, bug annihilation | Building new features, fixing bugs |
| Writer | Documentation, technical prose | API docs, README updates, guides |
| Reviewer | Code review, security sanctification | Pre-merge reviews, security audits |
| Orchestrator | Complex multi-step orchestration | Architecture changes, large refactors |
# Invoke the coder (default)
sylphx-flow "implement user dashboard"
# Summon the reviewer
sylphx-flow "review for security vulnerabilities" --agent reviewer
# Call upon the writer
sylphx-flow "document the API endpoints" --agent writer
# Unleash the orchestrator
sylphx-flow "refactor authentication system" --agent orchestratorConfigure enabled agents:
sylphx-flow settings
β Select: Agents & Default Agent# 1. Install the meta-CLI
npm install -g @sylphx/flow
# 2. First invocation (auto-setup)
sylphx-flow "create a REST API"
# Flow's first-time ritual:
β Detects environment
β Checks for Claude Code/OpenCode
β Installs missing dependencies (with your permission)
β Configures optimal settings
β Upgrades to latest versions
β Executes your command
# 3. Subsequent invocations
sylphx-flow "build something amazing"
# Flow's ongoing ritual:
β Auto-checks for updates
β Auto-upgrades if needed
β Executes with latest capabilities# Simple tasks (MEP in action)
sylphx-flow "add password reset"
sylphx-flow "fix the login bug"
sylphx-flow "optimize database queries"
# Specialized agents
sylphx-flow "review for security" --agent reviewer
sylphx-flow "write API documentation" --agent writer
sylphx-flow "refactor user service" --agent orchestrator
# File input for complex prompts
sylphx-flow "@detailed-requirements.md"
# Settings management
sylphx-flow settingsOne interface. All tools. Zero friction.
Flow intelligently detects and orchestrates:
- Claude Code - Anthropic's official CLI
- OpenCode - Open source alternative
- Future tools - Built for extensibility
# Same command, different environments
# Flow adapts automatically
# On machine with Claude Code
$ sylphx-flow "task" β Uses Claude Code
# On machine with OpenCode
$ sylphx-flow "task" β Uses OpenCode
# On machine with both
$ sylphx-flow "task" β Uses your preferred defaultMissing tools? Flow installs them for you.
# First time on a new machine
$ sylphx-flow "build something"
> No AI CLI detected.
> Which platform would you like to use?
1. Claude Code (Anthropic, recommended)
2. OpenCode (open source alternative)
> Your choice: 1
> Installing Claude Code...
> β Downloaded latest version
> β Configured settings
> β Installed MCP servers
> β Ready to execute
> Executing your command...Every session uses the latest capabilities.
Flow checks for updates before each execution:
$ sylphx-flow "task"
> Checking for updates...
> Updates found:
β’ Flow: 2.0.0 β 2.1.0
β’ Claude Code: 1.5.0 β 1.6.0
> Upgrading...
> β Flow updated (new features available)
> β Claude Code updated (performance improved)
> Executing with latest versions...Customize update behavior:
# Check updates without executing
sylphx-flow upgrade
# Auto-upgrade and execute
sylphx-flow upgrade --autoMinimal Effective Prompt: AI that truly understands context.
Flow automatically enriches every prompt with:
- π Your codebase patterns (via semantic search)
- π Best practices (via curated knowledge base)
- π₯οΈ Your environment (OS, Node version, git status)
- β° Current time and date
- π― Your preferred coding style
# You type 3 words
sylphx-flow "implement authentication"
# Flow expands to full context:
β Tech stack: TypeScript + React + Next.js (detected)
β Patterns: Your existing auth patterns (searched)
β Best practices: JWT, bcrypt, RBAC (knowledge base)
β Environment: macOS, Node 20, git clean (hooks)
β Time: 2025-11-23 (system)
# Result: Perfect, production-ready codeConfigure once. Perfect forever.
# Access settings nexus
sylphx-flow settings
# Configure:
β Agents (coder, writer, reviewer, orchestrator)
β Rules (core, code-standards, workspace)
β Output styles (silent, verbose, tutorial)
β MCP servers (grep, context7, playwright, github, notion)
β Providers (Anthropic, Kimi, Z.AI)
β Target platform (Claude Code, OpenCode)
# Settings apply to ALL subsequent executionsFlow respects your version control.
When Flow operates, it uses git update-index --skip-worktree to hide temporary changes:
# Before Flow execution
$ git status
β clean working directory
# During Flow execution (modifying .claude/config.json)
$ git status
β clean working directory # Changes hidden!
# After Flow execution
$ git status
β clean working directory # Restored automaticallyThis ensures:
- LLMs won't see or commit Flow's temporary settings
- Your version-controlled configs remain pristine
- Team collaboration stays clean
Set it and forget it. Flow works while you sleep.
# Continuous autonomous work
sylphx-flow "process all GitHub issues" --loop
# With polling interval
sylphx-flow "monitor and fix bugs" --loop 300 --max-runs 20
# Checks every 5 minutes, max 20 iterations
# How it works:
β Iteration 1: Fresh start
β Iteration 2+: Auto-continue with context
β Stop: Ctrl+C (graceful) or max runs reachedPerfect for:
- π Processing backlogs (issues, PRs, tickets)
- π Periodic monitoring and auto-fixing
- π§Ή Incremental refactoring
- π§ͺ Continuous test fixing
- π Documentation maintenance
Complex prompts deserve better than shell escaping.
# Load prompt from file
sylphx-flow "@task.txt"
# Absolute paths supported
sylphx-flow "@/path/to/detailed-requirements.md"
# Combine with other flags
sylphx-flow "@refactor-plan.md" --agent orchestrator --loopBenefits:
- π Write natural language (no shell escaping)
- π Version control your prompts
- π Reusable task definitions
- π No character limits
Find code by what it does, not what it's called.
# Search by functionality
sylphx-flow codebase search "user authentication logic"
# Works across languages
β Finds: authenticateUser() in TypeScript
β Finds: authenticate_user() in Python
β Finds: AuthenticateUser() in Go
# Multilingual queries
sylphx-flow codebase search "θηη¨ζΆη»ε
₯ε
ιθΌ―" # Chinese
sylphx-flow codebase search "γ¦γΌγΆγΌγγ°γ€γ³ε¦η" # Japanese
# Reindex after major changes
sylphx-flow codebase reindexPowered by StarCoder2 tokenization - 70+ languages supported.
Professional knowledge, zero maintenance.
# Search best practices
sylphx-flow knowledge search "react performance optimization"
sylphx-flow knowledge search "secure password hashing"
sylphx-flow knowledge search "microservices patterns"
# List all knowledge
sylphx-flow knowledge list
# Update to latest
sylphx-flow knowledge updateWhy curated?
- β Professionally maintained
- β Always up-to-date
- β Quality guaranteed
- β <100ms search
- β Zero cognitive load
# Rapid prototyping
sylphx-flow "create a blog with markdown support"
# Feature development
sylphx-flow "add OAuth login with GitHub"
# Bug fixing
sylphx-flow "fix the memory leak in user service"
# Refactoring
sylphx-flow "refactor to use dependency injection" --agent orchestrator# Code review automation
sylphx-flow "review PR #123 for security" --agent reviewer
# Documentation
sylphx-flow "document all API endpoints" --agent writer
# Consistency enforcement
# (Settings ensure everyone uses same rules/agents)
sylphx-flow settings
β Team admin configures standards
β All developers inherit settings# Architecture changes
sylphx-flow "migrate from REST to GraphQL" --agent orchestrator
# Security audits
sylphx-flow "audit for OWASP Top 10" --agent reviewer
# Performance optimization
sylphx-flow "optimize database queries" --loop
# Continuous maintenance
sylphx-flow "process GitHub issues" --loop 600 --max-runs 100- Installation & Setup - Complete setup guide
- Settings Guide - Master the settings nexus
- MEP Design Philosophy - Understand the paradigm shift
- Technical Architecture - Deep dive into Flow's internals
- CLI Commands - Full command reference
- Agent Framework - How agents work
- Loop Mode Guide - Autonomous execution mastery
| Platform | Status | Auto-Install | Loop Mode |
|---|---|---|---|
| Claude Code | β Full | β Yes | β Yes |
| OpenCode | β Full | β Yes | β³ Coming Soon |
| Cursor | β MCP | Manual | β³ Coming Soon |
- IDE Plugins - VSCode, IntelliJ, Vim
- CI/CD Integration - GitHub Actions, GitLab CI
- Cloud Execution - Remote AI orchestration
- Team Collaboration - Shared settings and knowledge
- Custom Agents - Create your own specialized agents
| Feature | Traditional AI CLIs | Sylphx Flow |
|---|---|---|
| CLI Count | Multiple (1 per tool) | One (rules them all) |
| Installation | Manual, fragmented | Automatic, unified |
| Updates | Manual per tool | Automatic per session |
| Settings | Per-project, scattered | Global, centralized |
| Context Awareness | Manual prompting | Automatic enrichment |
| Prompt Length | 500+ words | 3-10 words (MEP) |
| Git Integration | Accidental commits | Skip-worktree protection |
| Ctrl+C Handling | Crashes, manual recovery | Graceful restore |
| Agent System | Generic AI | Specialized experts |
| Knowledge Base | DIY or none | Professionally curated |
- β‘ 30x faster - Time to prompt (10s vs 5min)
- π 50x shorter - Prompt length (10 words vs 500)
- π― +25% - Context accuracy (95% vs 70%)
- π§ -80% - Cognitive load (minimal vs high)
- π Zero - Maintenance overhead (vs constant)
- π 5x faster - Team onboarding (days vs weeks)
Flow stands on the shoulders of giants:
- StarCoder2 - Revolutionary code tokenization
- MCP Protocol - Standard AI tool integration
- Anthropic Claude - Foundation model
- OpenCode - Open source alternative
- Open source community β€οΈ
This project uses the following @sylphx packages:
- @sylphx/biome-config - Shared Biome configuration
- @sylphx/bump - Version management and releases
- @sylphx/doctor - Project health checks and auto-fixes
MIT License - see LICENSE file for details.
β Star us on GitHub β’ π Read the Docs β’ π Report Issues
One CLI to rule them all.
Made with β€οΈ by Sylphx Ltd