recipe-saas-agent

What problem does it solve?

Deploying AI agents in multi-tenant SaaS environments requires strict isolation, cost control, safety filters, and reliable observability; this recipe provides a complete, production-ready configuration that enforces guardrails, per-user cost tracking, rate limiting, and auditability so tenants cannot interfere with one another or exceed budgets.

Core Features & Use Cases

• Prompt injection protection & PII masking: Detects and mitigates injection attempts and masks sensitive data before it reaches the model.
• Per-user cost budgets & rate limiting: Enforces per-session and daily budgets, token limits, requests-per-minute, and concurrency constraints.
• Behavioral contracts & circuit breakers: Limits tool usage, iterations, and output size while degrading gracefully on provider failures.
• A2A networking & observability: Exposes agents safely for internal service meshes, streams responses, and writes audit logs for compliance.
• Use Case: Host a per-request chat endpoint for a multi-tenant SaaS product that returns streaming answers, charges against user budgets, blocks unsafe content, and records every request for audit.

Quick Start

Create a per-request SaaS agent for user user123 on the pro plan and run it with the user's prompt to receive a guarded, cost-tracked, auditable response.