Further Reading
Resources for going deeper with AI agents.
Official Documentation
- OpenAI Function Calling — The underlying API that powers tool use
- Anthropic Tool Use — Claude’s approach to tools
- LiteLLM — The multi-provider library we use
Specifications
- Model Context Protocol (MCP) — Anthropic’s open standard for tool servers
Agent Frameworks
- LangGraph — LangChain’s graph-based agent framework
- OpenAI Agents SDK — OpenAI’s reference implementation
- Google Agent Development Kit — Google’s agent framework
Learning Resources
- Prompt Engineering Guide — Comprehensive guide to prompting
- LLM Course — Practical LLM learning path
Papers
- ReAct: Synergizing Reasoning and Acting — The reasoning + acting pattern
- Toolformer — Teaching LLMs to use tools
Extending AgentSilex
Ideas for building on top of this framework:
| Extension | Approach |
|---|---|
| Persistent memory | Store sessions in Redis/SQLite |
| Rate limiting | Wrap LiteLLM calls with rate limiter |
| Caching | Cache tool results with @lru_cache |
| Async support | Wrap sync code with asyncio.to_thread |
| Custom tracing | Extend the observability module |
The ~1000 lines of AgentSilex are meant to be read, understood, and modified. Fork it, break it, make it yours.