irandysousa/langgraph-llama-cpp-starter
🤖 Build intelligent, offline LLM agents with LangGraph and llama-cpp-python using this starter template for local, private tool-calling applications.
What's novel
🤖 Build intelligent, offline LLM agents with LangGraph and llama-cpp-python using this starter template for local, private tool-calling applications.
Code Analysis
6 files read · 3 roundsA CLI chatbot that loads a local GGUF model via llama-cpp-python, uses prompt-engineered JSON tool calling (regex-parsed from LLM output) to invoke basic math tools, orchestrated by a simple LangGraph state loop.
Strengths
The LangGraph state machine loop for tool calling is cleanly structured, and the prompt engineering for JSON tool calling with detailed system instructions shows practical understanding of working with local models that lack native function calling. The code is readable and well-documented with docstrings.
Weaknesses
No tests exist, the streaming default in _call() causes double output (prints tokens AND returns text), the tool calling is fragile regex-based JSON parsing, and the README is heavily misleading—claiming 'no programming knowledge needed' and describing a double-click application when it's a developer Python project requiring model downloads and environment setup.
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
13
Contributors
2
Files
10
Active weeks
4
Repository
Language
Python
Stars
1
Forks
0
License
MIT