What is Shannon?
Shannon is a production-grade AI agent orchestration platform designed to solve the critical challenges of deploying AI agents at scale:Cost Control
Hard budget limits, token tracking, and intelligent model selection often delivering 60–90% cost savings (workload‑dependent)
Reliability
Deterministic replay debugging, circuit breakers, and automatic degradation
Security
WASI sandboxing for code execution, OPA policy enforcement, multi-tenant isolation
Scalability
Temporal-backed distributed workflows, horizontal scaling, comprehensive observability
Key Features
Multi-Agent Orchestration
Shannon coordinates multiple AI agents using proven cognitive patterns like Chain-of-Thought (CoT), Tree-of-Thoughts (ToT), and ReAct, enabling complex task decomposition and parallel execution.Production-Ready
Built with production workloads in mind:- Temporal workflows for durable, deterministic execution
- WASI sandboxing for secure Python code execution
- Circuit breakers and failure protection
- Comprehensive observability with Prometheus metrics and OpenTelemetry tracing
Multi-Provider Support
Seamlessly switch between LLM providers:- OpenAI (GPT-5 family)
- Anthropic (Claude 3 Opus/Sonnet/Haiku)
- Google Gemini
- Groq, Azure OpenAI, Ollama, and more
Quick Start
Get Shannon running in under 10 minutes:Installation Guide
Set up Shannon with Docker Compose in one command
Architecture Overview
Shannon consists of four main components working together:Gateway
REST API layer with authentication and rate limiting
Orchestrator
Temporal-based workflow coordination and task routing
Agent Core
Secure execution layer with WASI sandboxing
LLM Service
Multi-provider LLM gateway with intelligent caching
Use Cases
Shannon excels at:- Complex Task Automation: Break down complex tasks into manageable subtasks with automatic orchestration
- Research & Analysis: Coordinate multiple agents for comprehensive research and synthesis
- Code Generation: Secure Python code execution in WASI sandbox
- Multi-Step Workflows: Durable workflows that survive failures and can be replayed for debugging