Examples Guide

The examples in this repository are runnable research-oriented scripts. They are designed to show not only API usage, but how the library fits into realistic experimental workflows. Each example lists dependencies, expected scope, and the primary concept it demonstrates.

Note

Some local LlamaCppServerLLMClient examples intentionally use Qwen3-4B GGUF configs, which can exceed available RAM on smaller machines. If you want a lighter local starting point, swap in a smaller model or begin with Ollama Local Client, which documents the lighter Ollama defaults.

Featured Examples

Direct LLM Call

One-step participant execution with a configured backend client.

Requires: base install + reachable backend endpoint Runtime: short Teaches: baseline participant setup, request execution, structured output handling

Multi-Step JSON Tool Calling Agent

Iterative tool-using execution with explicit action/observation loops.

Requires: base install Runtime: short to medium Teaches: tool-routing behavior, multi-step control, inspectable intermediate state

Debate Pattern

Role-based multi-agent coordination with adjudication workflow structure.

Requires: base install Runtime: medium Teaches: orchestration patterns, delegate coordination, traceable multi-role reasoning

MCP Minimal

Small end-to-end MCP-backed tool integration example.

Requires: mcp-compatible server/runtime setup Runtime: medium Teaches: external tool connectivity, MCP source wiring, runtime safety boundaries

Deterministic runs for tests are provided by tests/example_monkeypatch/sitecustomize.py when DRA_EXAMPLE_LLM_MODE=deterministic is set.

Full Catalog

• Agent Examples
  • Direct LLM Call
  • Direct LLM Compiled Execution
  • Direct LLM With Pinned Skills
  • Multi Step Code Tool Calling Agent
  • Multi Step Direct LLM Agent
  • Multi Step JSON Tool Calling Agent
  • Multi Step JSON With Memory
  • Multi Step JSON With Skills
  • Prompt Workflow Agent
  • Seeded Random Baseline Agent
  • VS Code Hello World
• Workflow Primitive Examples
  • Workflow Delegate And Memory Steps
  • Workflow Diagram Generation
  • Workflow Model Step Design Tradeoff
  • Workflow Prompt Mode
  • Workflow Runtime
  • Workflow Runtime Loop Step
  • Workflow Schema Mode
• Pattern Examples
  • Coordination Patterns
  • Debate Pattern
  • Nominal Team
  • Plan Execute
  • Propose Critic
  • Rag
  • Ralph Loop
  • Router Delegate
  • Simulated Annealing
  • Tree Search
  • Two Speaker Conversation
• Client Examples
  • Anthropic Service Client
  • Gemini Service Client
  • Groq Service Client
  • Llama CPP Server Client
  • MLX Local Client
  • Ollama Local Client
  • OpenAI Compatible HTTP Client
  • OpenAI Service Client
  • SGLang Server Client
  • Transformers Local Client
  • vLLM Server Client
• Model Selection Examples
  • Local
  • Remote
• Tool Examples
  • MCP Minimal
  • Multi Source Tool Usage
  • Script Tools / Repo Quickscan
  • Script Tools / Rubric Score
• Optimization Examples
  • Multi Step JSON Tool Calling 1d Optimization