Saifuban adds daily API budgets to cap runaway AI agent spending

A developer built a daily API spending cap into Saifuban, a guard tool for AI agents, after worrying about agents stuck in loops running up large bills overnight. Users store their provider API key in Saifuban and receive a proxy key, which the agent uses so that all requests route through the service. Saifuban tallies each request's cost from the usage data in responses and, once a per-agent daily budget is reached, returns a 403 error instead of forwarding further calls until the next day. The cap supports OpenAI, Anthropic, Google Gemini, and OpenRouter, with costs calculated from token counts and price tables for all but OpenRouter, which reports actual cost. Requests already in flight when the limit is hit may push spending slightly over budget, and users must trust the tool with their real API keys, which are stored encrypted.