Budget-Capped AI Agents in n8n: OpenAI Client Runs via Webhook

Run budget-capped AI agents for multiple clients in n8n—via a secure webhook

This n8n automation lets your app trigger an OpenAI-powered AI agent for multiple paying clients while enforcing per-client monthly token budgets, pause controls, and safe concurrency using an AI usage ledger.

What this workflow does

• Your app POSTs client_id and a prompt to the workflow’s webhook.
• If the client is unknown or paused, the webhook returns 403. If the client has no budget left, it returns 429.
• To prevent overspending, the workflow writes a worst-case token reservation to a run-specific ledger row in ai_usage_ledger.
• Before executing, the ledger is read again; if other parallel runs already consumed budget, the run backs off.
• The AI Agent runs with Max Iterations and a token limit taken from the workflow’s Settings nodes.
• After each model call, the workflow uses the n8n API to read recorded prompt/completion token usage, then books tokens to the correct client and returns the answer along with a usage summary.
• You receive an email alert at the configured threshold and when a client can’t afford another run.
• Hourly stuck reservations expire; monthly totals are calculated per client and old ledger rows are deleted. A GET endpoint returns usage per client.
• If the workflow can’t read/write the Data tables, it refuses the run with 503.

Use cases

• SaaS agencies running one n8n AI agent for multiple customers with monthly token limits.
• Freelancers offering AI automation retainers where each client’s usage is tracked and capped.
• Multi-tenant internal tools that need webhook-based AI execution and auditable usage reporting.

Technical details

• Data tables: ai_clients, ai_usage_ledger
• Nodes/integrations: webhook, Data table, if, set, code, n8n API (read token usage), OpenAI (chat model), SMTP (email alerts)
• Security: Header Auth on both webhooks
• Setup: configure Settings nodes (keep Save execution progress enabled), add client rows, and publish the workflow.