Detailed Introduction
AgentSilex is a developer-focused, lightweight agent framework designed to be transparent, minimal, and hackable. The core implementation is intentionally compact (around 300 lines), making it easy to read, understand, and modify. Built on top of LiteLLM and compatible with many model providers, AgentSilex enables seamless switching between OpenAI, Anthropic, Google Gemini, local LLMs, and more.
Main Features
- Transparent architecture: readable code with no hidden complexity, ideal for education and auditing.
- Minimal core: a small codebase that reveals the full execution flow at a glance.
- Hackable: designed for forks and customization, suitable for production prototypes and research.
- Universal model support: switch providers with minimal code changes.
Use Cases
- Education and research: a compact reference implementation for courses and experiments.
- Enterprise PoC: a controllable foundation for building customized agent systems.
- Developer learning: understand tool calling, session management, and multi-agent handoffs.
Technical Features
- Tool calling with
@tooldecorator and type-safe parameter extraction. - Session and conversation history management with multi-agent handoff support.
- Built-in OpenTelemetry observability for tracing tool calls and execution flows.
- Streaming responses and event-driven execution model for improved user experience.