llm-framework

Minimal Python library for building LLM-powered agents with a low dependency footprint.

• core/ needs only httpx and python-dotenv. Always available.
• extensions/ are optional extras. Pull only what you need.
• Every feature targets the "goldilocks" zone: use a library where it saves meaningful complexity, write from scratch where it adds unnecessary dependencies.

---

Where to go next

• Installation: Learn how to install the core library and its optional extensions.
• Quickstart: Build your first agent in less than 15 lines of code.
• Patterns: Architecture guidance for choosing tools/ or mcp_servers/.

---

API Reference Overview

• Core — LLMClient, Agent, Orchestrator, HistoryBuffer, @tool, @cached_tool
• Extensions — MCPClient, MCPManager, MCPServer, MCPContext, MemoryStore, RAGStore, auth gate, OIDCAuthProvider, guardrails, vector stores
• Tools — Filesystem, shell, web fetch, calculator, clock, memory
• Configuration — Config dataclass