Fast-CrewAI

Drop-in performance acceleration for CrewAI - Get up to 34x faster serialization, 17x faster tool execution, and 11x faster database search with zero code changes to your existing CrewAI projects.

Fast-CrewAI is authored by Dipankar Sarkar and maintained by Neul Labs to bring production-ready acceleration to CrewAI workloads.

🚀 Quick Start

Get your CrewAI code running faster in seconds:

# Install the acceleration layer (using uv - recommended)
uv add fast-crewai

# Or with pip
pip install fast-crewai

# Add one line to your existing CrewAI code
import fast_crewai.shim  # Add this before importing CrewAI
from crewai import Agent, Task, Crew  # Your existing code remains unchanged!

That's it! Your existing CrewAI code now benefits from Rust-powered acceleration.

🏎️ What Gets Faster

| Component | Performance Boost | What This Means | |-----------|-------------------|-----------------| | Serialization | 🚀 34x faster | Agent message serialization using serde | | Tool Execution | 🚀 17x faster | Result caching and JSON validation | | Database Search | 🚀 11x faster | FTS5 full-text search with BM25 ranking | | Memory Storage | ✅ TF-IDF search | Semantic search using cosine similarity | | Task Execution | ✅ Dependency tracking | Topological sort and parallel scheduling |

See BENCHMARK.md for detailed performance data.

🤝 How It Works

Fast-CrewAI uses smart monkey patching that seamlessly integrates with your existing CrewAI code:

# Your existing CrewAI code - no changes needed!
from crewai import Agent, Task, Crew  # This automatically gets accelerated!

# Just add this line once at the top of your main file
import fast_crewai.shim  # Activates acceleration for all CrewAI components

The acceleration happens automatically through dynamic inheritance - your tools, tasks, and other components get wrapped with performance improvements while keeping 100% API compatibility.

📦 Installation

Get Started Quickly

# Using uv (recommended)
uv add fast-crewai

# Or with pip
pip install fast-crewai

For Development

# Install uv if you haven't already
curl -LsSf https://astral.sh/uv/install.sh | sh

# Clone and setup
git clone https://github.com/neul-labs/fast-crewai.git
cd fast-crewai
uv sync --dev

# Build Rust extensions for maximum performance
uv run maturin develop --release

🧪 Testing & Verification

Verify Fast-CrewAI is working with these tests:

# Run all unit tests
make test

# Compare performance (with/without acceleration) - runs full CrewAI workflows
make test-comparison

# Run extensive performance tests (1000 iterations)
make test-comparison-extensive

# Full compatibility check with CrewAI
make test-compatibility

# Run internal benchmarks
make benchmark

The test-comparison command runs complete CrewAI workflows that engage all acceleration components:

• Memory acceleration: Through CrewAI's memory systems (RAG storage, short/long-term memory)
• Database acceleration: Through SQLite operations for memory persistence
• Tool execution: Through any tools used in the workflow (hooks enabled)
• Task execution: Through CrewAI's task processing system (hooks enabled)

For example, on basic workflows the comparison typically shows around 13% performance improvement:

==================================================
Workflow Type: basic
Iterations: 3
Baseline (without Fast-CrewAI): 27.88s
Accelerated (with Fast-CrewAI): 24.50s
Time Saved: 3.38s
Performance Improvement: 1.13x faster
Percent Improvement: 13.00%
==================================================

Higher improvements (2-5x) are typically seen with memory-intensive and database-heavy workflows where Fast-CrewAI's Rust acceleration provides greater benefits.

⚙️ Configuration

Fine-tune your acceleration with environment variables:

# Enable all acceleration (default)
export FAST_CREWAI_ACCELERATION=1

# Disable acceleration (for debugging)
export FAST_CREWAI_ACCELERATION=0

# Control specific components
export FAST_CREWAI_MEMORY=true    # Memory acceleration
export FAST_CREWAI_TOOLS=false    # Tool acceleration
export FAST_CREWAI_DATABASE=true  # Database acceleration

📊 Performance Benchmarks

Latest benchmark results (see BENCHMARK.md for full details):

Performance Improvements

| Component | Improvement | Details | |-----------|-------------|---------| | Serialization | 🚀 34.5x faster | 80,525 ops/s (Rust) vs 2,333 ops/s (Python) | | Tool Execution | 🚀 17.3x faster | Result caching with configurable TTL | | FTS Database Search | 🚀 11.2x faster | 10,206 ops/s (FTS5) vs 913 ops/s (LIKE) | | Database Query | 🚀 1.3x faster | Connection pooling with r2d2 | | Memory Storage | ✅ Comparable | TF-IDF semantic search |

Memory Usage

| Component | Python | Rust | Savings | |-----------|--------|------|---------| | Tool Execution | 1.2 MB | 0.0 MB | 99% less | | Serialization | 8.0 MB | 3.4 MB | 58% less | | Database | 0.1 MB | 0.1 MB | 31% less |

Key Features by Component

Database (FTS5)

• Full-text search with BM25 relevance ranking
• Automatic index synchronization via triggers
• Connection pooling for concurrent access

Tool Executor

• Result caching with TTL expiration
• Fast JSON validation using serde
• Execution statistics tracking

Task Executor

• Dependency tracking with topological sort
• Cycle detection for circular dependencies
• Parallel task scheduling via Tokio

Serialization

• serde-based JSON serialization (34x faster)
• Lower memory footprint (58% less)

🛠️ Development & Testing

The project includes comprehensive tools for development and performance analysis:

# Run all tests
make test

# Run performance comparisons
make test-comparison                 # Quick comparison
make test-comparison-extensive      # 1000 iterations for detailed analysis

# Run compatibility tests
make test-compatibility             # Verify with latest CrewAI

# Run internal benchmarks
make benchmark                      # Test components in isolation

📚 Learn More

• Architecture: How the acceleration system works under the hood
• API Reference: Complete API documentation
• Configuration: Fine-tuning Fast-CrewAI for your needs
• Performance: Detailed benchmarking and optimization
• Troubleshooting: Common issues and solutions

For complete documentation, visit the docs/ directory.

📋 Requirements & Compatibility

Python Version Support

| Python Version | Status | Notes | |----------------|--------|-------| | Python 3.10 | ✅ Supported | Fully tested | | Python 3.11 | ✅ Supported | Fully tested | | Python 3.12 | ✅ Supported | Fully tested | | Python 3.13 | ✅ Supported | Fully tested | | Python 3.9 | ⚠️ Limited | May work but not tested in CI |

Dependencies

• CrewAI: Any recent version supported
• Rust toolchain: Optional, for native extensions (provides maximum performance)
  • PyO3 0.23+ is used for Python-Rust bindings
  • Falls back to pure Python if Rust extensions aren't available

Platform Support

| Platform | Status | |----------|--------| | Linux (x86_64) | ✅ Fully supported | | macOS (x86_64) | ✅ Fully supported | | macOS (ARM64/M1) | ✅ Fully supported | | Windows (x86_64) | ✅ Supported |

🤝 Contributing

Contributions are welcome! Check out our development guide for setup instructions and coding standards.

📄 License

MIT License - see LICENSE file for details.

👥 Maintainers

• Author: Dipankar Sarkar ([email protected])
• Organization: Neul Labs

🔧 Support

• Issues: GitHub Issues
• Discussions: GitHub Discussions

Fast-CrewAI is in active development. Memory and database acceleration are production-ready, with ongoing optimization for tools and tasks.