Skip to content

Demos & Live Proofs

CRPv6 ships with a set of runnable demos that prove the protocol works against real local language models. Each demo is a plain Python script; no managed cloud or paid API key is required.


Quick start: run the live SLM proof

  1. Start LM Studio and load any 8B instruct model.
  2. Run:
export CRP_LMSTUDIO_URL=http://localhost:1234/v1
export CRP_LMSTUDIO_MODEL=meta-llama-3.1-8b-instruct
python examples/crp_demos/live_crp_slm_proof.py

The script compares raw LLM output with CRPv6 output on four tasks:

  • Single tool - execute one tool and return a natural answer.
  • Tool chain - run two tools in sequence with real intermediate results.
  • RAG retrieval - search a local knowledge base and synthesise an answer.
  • Long-form report - produce a structured, non-repetitive multi-paragraph report.

For a step-by-step video guide, see the Video Demo Guide.


All demos

Demo File What it proves
Live SLM proof examples/crp_demos/live_crp_slm_proof.py Raw LLM vs. CRPv6 on 4 tasks with full governance output.
LLM vs. CRP examples/crp_demos/live_llm_vs_crp.py One-question side-by-side comparison.
Agent test harness examples/crp_demos/live_agent_test_harness.py 4-case pass/fail harness against a live SLM.
Real-world API research agent examples/agents/real_world_research_agent.py SLM agent calling live weather, crypto, and Wikipedia APIs (no API keys).
Real-world API proof examples/crp_demos/live_real_world_api_proof.py Side-by-side raw LLM vs. CRPv6 on live public APIs.
Weather agent examples/agents/slm_weather_agent.py SLM-ready agent with weather tools.
RAG agent examples/agents/slm_rag_agent.py SLM-ready RAG over an in-memory corpus.
Mock weather agent examples/agents/weather_agent.py Same agent, mock provider, works without a model.
Mock RAG agent examples/agents/rag_agent.py Same RAG agent, mock provider.
GDPR DSR agent examples/agents/gdpr_dsr_agent.py Sensitive workflow with checkpoint and audit.
Report agent examples/agents/report_agent.py Scan → summarise → report pipeline.

What is real vs. hardcoded

Real Hardcoded
Model output, both raw and CRP Weather tool return value (22°C and sunny) so the demo works offline
CRP governance metadata Pass/fail assertions in the harness
Tool selection and execution Mock provider responses in non-live templates
RAG search results Sample corpus documents

The protocol's value is in the governance wrapper and tool loop, not in the sample data. Hardcoded sample data is acceptable; hardcoded governance would defeat the point.


Run the test suite

The non-live test suite proves the protocol without requiring API keys:

pytest tests/ -q --tb=short

Expected: 3232+ tests pass.

For a full public test/demo runbook, see the Public Demo & Test Guide.