n8n Log AI Agent Runs with Hash-Chained Receipts & OpenAI

Run an OpenAI AI agent with tamper-evident logging—every request gets a hash-chained receipt

This n8n workflow exposes secure webhook endpoints that run an OpenAI-powered agent per client, then records each run into an n8n Data Table using a SHA-256 hash chain. It also generates daily integrity reports and anchors and provides a verification endpoint for clients.

What this workflow does

• Receives webhook runs via POST with client_id and prompt, validating client and prompt rules and rejecting invalid requests immediately.
• Executes a LangChain AI agent using OpenAI Chat (with an optional Calculator tool) to produce an answer or capture an agent failure.
• Creates a logged receipt by hashing the prompt and answer (optionally storing clipped text), counting model/tool calls, and tagging the outcome as ok, refused, or error.
• Appends a new hash-chained entry into the agent_run_log table by linking to the latest entry’s hash, computing the next entry_hash, and confirming canonical sequence (with short backoff retries on collisions).
• Returns the result with integrity proof including seq and entry_hash. If reliable logging fails, it withholds the agent output and returns HTTP 503.
• Notifies the owner and verifies daily: a scheduled run recomputes and verifies all clients’ hash chains in agent_run_log, stores anchors in agent_run_anchors, and emails a report (or failure notice).

Use cases

• SaaS operators needing auditable AI responses for compliance and dispute resolution.
• Automation engineers implementing tamper-evident agent logs for multi-tenant n8n apps.
• Teams that want client-accessible verification of each run via a dedicated endpoint.

Technical details

• n8n nodes/tools: webhook, if, set, code, wait, switch
• AI stack: LangChain AI agent with OpenAI Chat and optional Calculator tool
• Storage: agent_run_log and agent_run_anchors n8n Data Tables
• Integrity: SHA-256 hash chaining, sequence verification, daily recomputation/anchoring