Examples¶

Nine numbered examples — all runnable offline with scripted providers — plus two applications that need real providers.

Run any of them from the repository root:

git clone https://github.com/apavanello/toolloop && cd toolloop
uv sync --extra dev
uv run python examples/01_first_agent.py

The guided tour¶

#	Example	You will learn
01	`01_first_agent.py`	the three moving parts: provider, tools, agent loop
02	`02_custom_tools.py`	`@tool`, pydantic validation, the auto-repair loop
03	`03_control_and_hooks.py`	`APPROVE` vs `BYPASS`, deny/rewrite calls, observability
04	`04_subagent.py`	delegation with an isolated context
05	`05_context_window.py`	truncation + compaction, pluggable `token_counter`
06	`06_mcp_tools.py`	MCP servers as tools (spawns its own server — offline)
07	`07_parallel_and_streaming.py`	`max_parallel_calls` wall-clock, `stream()` + `on_delta`
08	`08_sessions_and_checkpoints.py`	save/resume a conversation, incremental checkpoints
09	`09_production_hardening.py`	retries, timeout, rate limiting, usage, `run_sync` — all sync

coding_agent.py — a minimal coding harness: standard toolset, ControlMode.APPROVE with console_approver, dev logging, retries/timeout and a crash-resume checkpoint.

OPENAI_API_KEY=... uv run --extra openai python examples/coding_agent.py "task"

repo_summarizer.py — a repository summarizer on OpenRouter with a custom tool, an on_step progress hook and structured output validated by pydantic.

OPENROUTER_API_KEY=sk-or-... uv run --extra openai python examples/repo_summarizer.py .