Fast-CrewAI

How it works

A Rust layer that hides behind CrewAI's own API

Fast-CrewAI never asks you to learn a new API. It slips underneath CrewAI at import time, routes the expensive paths through Rust, and returns control to your unchanged code.

Your application code
Agents · Tasks · Crews — unchanged
CrewAI public API
100% compatible surface (101 tests)
Fast-CrewAI shim
Dynamic inheritance patches hot paths at import time
PyO3 boundary
Zero-copy calls into compiled Rust
Rust engines
serde · SQLite FTS5 + BM25 · tool cache · r2d2 pool

Requests flow down; accelerated results flow back up. Only the shaded layers are Fast-CrewAI — everything above is stock CrewAI and your code.

The four-step lifecycle

01

You import the shim first

import fast_crewai.shim runs before any crewai import. This is the only line you add — it registers the acceleration layer before CrewAI classes are constructed.

02

Dynamic inheritance patches the hot paths

The shim subclasses the relevant CrewAI components and swaps their serialization, memory, and tool-execution internals for Rust-backed implementations — without altering the public API.

03

PyO3 bridges Python and Rust

Calls on the accelerated paths cross into compiled Rust via PyO3. serde handles JSON, FTS5 handles memory search, and cached tool results skip repeated execution.

04

Your code runs unchanged, just faster

Your Agents, Tasks, and Crews keep the exact CrewAI API. If a native wheel is missing for a platform, the layer falls back to pure Python so nothing breaks.

See it in code on the features page, or read Rust/Python interop with PyO3.

Ready to make CrewAI faster?

Talk to the team that wrote the acceleration layer. We take on performance audits, full system builds, and retained engineering.