OnceOnly Docs

Execution safety for AI agents — idempotency, budgets, tool control, audit.

Prevent duplicate actions

Use idempotency keys + dedupe windows to guarantee “at-most-once” execution for tools and workflows.

Control tools safely

Allow/deny tools, set per-tool call caps and rate limits, and keep risky tools gated by policy.

Enforce budgets & limits

Daily spend limits, per-agent budgets, action caps — stop runaway agents before they burn money.

Audit everything

Every action logged with context for debugging, compliance, and post-incident review.

API Reference

Start with Tools, Policies, and Agents endpoints to integrate OnceOnly into your stack.

Python SDK

Use the Python SDK to register tools, apply policies, and run agent-safe workflows.

Recommended path