Configuration

Configure models, retries, and runtime knobs in ~4 minutes. Requires Python 3.10+ and pip install alloy-ai.

---

Quick Start

export OPENAI_API_KEY=...
export ALLOY_MODEL=gpt-5-mini

Precedence

1. Per‑call overrides (e.g., ask(..., model=...))
2. configure(...) and context scopes
3. Environment variables (ALLOY_*)
4. Built‑in defaults (model="gpt-5-mini", max_tool_turns=10)

All Settings (env)

Name	Type	Default	Description
ALLOY_MODEL	str	gpt-5-mini	Model ID; determines provider (e.g., gpt-5-mini, claude-…, gemini-…, ollama:<model>)
ALLOY_TEMPERATURE	float	None	Sampling temperature; provider-specific handling
ALLOY_MAX_TOKENS	int	None	Cap on tokens for responses; some providers require this (Anthropic)
ALLOY_DEFAULT_SYSTEM	str	None	Default system prompt; alias: ALLOY_SYSTEM
ALLOY_RETRY	int	None	Retry count for transient failures
ALLOY_MAX_TOOL_TURNS	int	10	Max tool-call turn iterations in a single command run
ALLOY_AUTO_FINALIZE_MISSING_OUTPUT	bool	true	Issue one follow-up turn (no tools) to produce final structured output when missing
ALLOY_EXTRA_JSON	JSON object	{}	Provider-specific extras, merged into request (advanced)

Programmatic

from alloy import configure

configure(model="gpt-5-mini", temperature=0.2)