Lion Cipher Lab
Build your first AI harness: an LLM that triages log lines with a tool
- AI Security
- Intermediate
- ~60 min
- Free · no account required
Why it matters
A language model on its own is just a text box — you paste something in, you read something out. It becomes useful software the moment you wrap it in a harness: a program that feeds it inputs in a loop, gives it a tool or two it can call, and — the part everyone skips — checks its answers before doing anything with them. That wrapper is what every real AI feature is made of, from alert triage to code review bots. This lab has you build the smallest honest one end to end, on your own machine for free, so you understand it from the inside instead of from a vendor's marketing. Just as…
What you'll learn
- building an AI harness (a program that wraps an LLM)
- calling a local model from Python over Ollama's HTTP API
- structured output with a JSON schema
- the tool-use loop: letting a model request a function call
- validating and guardrailing model output before trusting it
- confidence thresholds and human-in-the-loop routing
- judging when an AI harness is the right tool and when it is not
Steps in this lab
- What an AI harness is — and when it's the right tool
- Install Ollama and a small model
- Define the tool and the synthetic logs
- Ask the model for STRUCTURED output, not prose
- The guardrail: validate before you trust
- The loop: tool call, confidence threshold, human-in-the-loop
- Prove the guardrail works: try to break your own harness
- Make it better — and know when NOT to reach for this at all