Skip to content

Traigent Examples Guide

What lives in examples/, how to run it, and where to start.

The canonical written guides live in docs/examples/. The browser gallery and inline demo assets live in examples/gallery/.

Layout (trimmed to what matters)

examples/
|- core/                    # Start here
|  |- simple-prompt/       # Minimal prompt tuning
|  |- rag-optimization/    # RAG toggle
|  |- few-shot-classification/
|  |- multi-objective-tradeoff/
|  |- prompt-ab-test/
|  |- structured-output-json/
|  |- token-budget-summarization/
|  |- safety-guardrails/
|  |- tool-use-calculator/
|  |- prompt-style-optimization/
|  `- chunking-long-context/
|
|- advanced/                # Patterns and deep dives
|  |- execution-modes/
|  |- results-analysis/
|  |- ai-engineering-tasks/
|  |- ragas/
|  `- metric-registry/
|
|- integrations/           # CI/CD and Bedrock samples
|- datasets/               # Shared evaluation data
|- gallery/                # Browser gallery and inline demo assets
`- templates/              # Example boilerplates

Core examples at a glance

Example Optimizes Run (mock mode)
simple-prompt Model, temperature, prompt style python examples/core/simple-prompt/run.py
rag-optimization RAG toggle, model, top_k TRAIGENT_MOCK_LLM=true python examples/core/rag-optimization/run.py
few-shot-classification Example count/strategy TRAIGENT_MOCK_LLM=true python examples/core/few-shot-classification/run.py
multi-objective-tradeoff Accuracy, cost, latency TRAIGENT_MOCK_LLM=true python examples/core/multi-objective-tradeoff/run_anthropic.py
prompt-ab-test Prompt variants TRAIGENT_MOCK_LLM=true python examples/core/prompt-ab-test/run.py
structured-output-json Schema validation TRAIGENT_MOCK_LLM=true python examples/core/structured-output-json/run.py
token-budget-summarization Budgeted summaries TRAIGENT_MOCK_LLM=true python examples/core/token-budget-summarization/run.py
safety-guardrails Moderation tuning TRAIGENT_MOCK_LLM=true python examples/core/safety-guardrails/run.py
tool-use-calculator Tool calling TRAIGENT_MOCK_LLM=true python examples/core/tool-use-calculator/run.py
prompt-style-optimization Tone/voice tuning TRAIGENT_MOCK_LLM=true python examples/core/prompt-style-optimization/run.py
chunking-long-context RAG chunking TRAIGENT_MOCK_LLM=true python examples/core/chunking-long-context/run.py

How to run

# Install from repo root (includes example deps)
pip install "traigent[recommended]"

# Optional: browse the gallery in a browser
python -m http.server 8000
# Open http://localhost:8000/examples/

# Auto-mocks when ANTHROPIC_API_KEY is unset
python examples/core/simple-prompt/run.py

# Real APIs (set keys you need)
export ANTHROPIC_API_KEY="your-key" # pragma: allowlist secret
export OPENAI_API_KEY="your-key" # pragma: allowlist secret
python examples/core/multi-objective-tradeoff/run_openai.py

Pick by goal

  • Lower cost/latency: multi-objective-tradeoff, token-budget-summarization, execution-modes (advanced).
  • Improve accuracy: few-shot-classification, prompt-style-optimization, ai-engineering-tasks/p0_few_shot_selection.
  • Safer outputs: safety-guardrails, ai-engineering-tasks/p1_safety_guardrails.
  • Structured JSON: structured-output-json, ai-engineering-tasks/p0_structured_output.
  • Tool calling: tool-use-calculator.
  • Long docs: chunking-long-context, ai-engineering-tasks/p0_context_engineering.

Quick pattern

import traigent

@traigent.optimize(
    eval_dataset="examples/datasets/simple-prompt/evaluation_set.jsonl",
    configuration_space={"model": ["claude-haiku-4-5-20251001"], "temperature": [0.0, 0.7]},
    objectives=["accuracy"],
    injection_mode="seamless",
)
def summarize(text: str) -> str:
    config = traigent.get_config()
    return f"Summary for: {text} | model={config['model']}"

Need a deeper map? Check Start Here for sequencing or Learning Roadmap for a multi-week path.