MagiC
MagiC is an AI-native framework for managing fleets of AI agents, providing an LLM gateway, prompt management, agent memory, and worker orchestration. It solves the infrastructure challenges of building AI applications, such as multi-provider routing, cost tracking, and workflow execution. It is for developers building and operating AI agent systems.
✨ Key features
- Multi-provider LLM gateway with routing and cost tracking
- Versioned prompt templates with A/B testing
- Agent memory with conversation history and vector recall
- DAG workflow orchestration with parallel execution and failure handling
- Worker registry with capability-based routing
- Built-in rate limiting and authentication
🎯 Use cases
- Build a fleet of specialized AI workers (e.g., search, summarization, analysis)
- Orchestrate multi-step AI workflows with dependencies and parallel execution
- Manage and route tasks to the best worker based on capabilities and cost
- Track and control AI spending with budget alerts and auto-pause
- Add shared knowledge and semantic search to AI agents
📦 Installation
🧰 Requirements: Requires Go 1.25+ and Python 3.11+ for source builds; Docker available. Optional PostgreSQL for production storage and API key for authentication.
# Option A: pip install (fastest)
pip install magic-ai-sdk
# Option B: From source
# Prerequisites: Go 1.25+, Python 3.11+
git clone https://github.com/kienbui1995/magic.git
cd magic
cd core && go build -o ../bin/magic ./cmd/magic && cd ..
./bin/magic serve
cd sdk/python && pip install -e . && cd ../..
docker run -p 8080:8080 kienbui1995/magic:latestdocker build -t magic .
docker run -p 8080:8080 magic🚀 Usage
# worker.py
from magic_ai_sdk import Worker
worker = Worker(name="HelloBot", endpoint="http://localhost:9000")
@worker.capability("greeting", description="Says hello")
def greet(name: str) -> str:
return f"Hello, {name}! Managed by MagiC."
worker.register("http://localhost:8080")
worker.serve()
python worker.py
Submit a task:
curl -X POST http://localhost:8080/api/v1/tasks \
-H "Content-Type: application/json" \
-d '{"type":"greeting","input":{"name":"World"}}'
❓ FAQ
How do I install MagiC?
You can install via pip (magic-ai-sdk), build from source (Go 1.25+, Python 3.11+), or use Docker. See the Quick Start section.
What programming languages are supported for workers?
The core is Go, and the SDK is Python, but workers can be in any language as they communicate via HTTP. The README shows examples in Python, Node, and Go.
How does MagiC handle cost tracking?
It automatically tracks tokens and model costs per request, per worker, and provides budget alerts and auto-pause via the Cost Controller.
Can I use MagiC with CrewAI or LangGraph?
Yes, MagiC is designed to power them. Your CrewAI agent becomes a MagiC worker with LLM routing, prompt management, and memory built in.
📊 Repository
🤖 Overview, features, install steps and FAQ were generated from the project's README on Sep 4, 2026. Always check the original source before running commands.