examples

Examples

Examples organized by concept, from simple to advanced.

Getting Started

Example	What it demonstrates
greeter/	Minimal single-agent setup—no tools, just a friendly prompt

Custom Tools

Example	What it demonstrates
calculator/	Python tools for math operations (factorial, fibonacci, add, multiply)
data_processor/	Data transformation tools (CSV formatting, statistics)

Custom Providers

Example	What it demonstrates
custom_provider/	Register a custom provider for use with LLM_DO_MODEL

File Operations & Approvals

Example	What it demonstrates
approvals_demo/	Approval workflow for file system writes—human confirmation for sensitive actions
file_organizer/	LLM semantic decisions + Python sanitization tools for file renaming
code_analyzer/	Read-only shell access with pre-approved commands (ls, find, grep, wc)

Progressive Stabilizing

The pitchdeck examples show the same task at different stabilization levels—a concrete demonstration of "extend with LLMs, stabilize with code."

Example	What it demonstrates
pitchdeck_eval/	All LLM: multi-agent delegation, PDF attachments, orchestrator pattern
pitchdeck_eval_stabilized/	Extracted tools: helper toolsets for common file operations
pitchdeck_eval_code_entry/	Python entry: orchestration in code, agents handle reasoning
pitchdeck_eval_direct/	Direct API: three abstraction levels without CLI

Recursive Patterns

Example	What it demonstrates
recursive_summarizer/	Hierarchical summarization—split, summarize chunks, merge recursively
recursive_task_decomposer/	Task breakdown across recursion depths with structured output

Web & Research

Example	What it demonstrates
web_searcher/	Server-side tools (native web_search) for live data
web_research_agent/	Multi-agent research pipeline: search → extract → consolidate → report

Orchestration Patterns

Example	What it demonstrates
orchestrating_tool/	Tool that orchestrates agents—encapsulates multi-agent pipelines in Python

Advanced Patterns

Example	What it demonstrates
whiteboard_planner/	Multi-modal input: reads whiteboard images, generates project plans
rlm_repl/	Persistent Python REPL context for querying large data

Running Examples

From the project root:

# Set your API key
export ANTHROPIC_API_KEY="sk-ant-..."

# Run any example
llm-do examples/greeter/project.json "Hello!"
llm-do examples/calculator/project.json "What is 10 factorial?"
llm-do examples/pitchdeck_eval/project.json

Some examples require input files (pitchdeck PDFs, whiteboard images). Check each example's project.json for expected inputs.