Audit hook and retention policy in eve

An eve hook that writes one hash-chained SQLite row per settled tool call, approval request and approval decision, keyed by stream position so a replay adds nothing. An eval approves one send and denies another, with a scripted mock model, and checks the rows, including who cleared or refused each call.

Back to the exercise

Before you start

Before you start:Install Claude Code Durable agents with eve Get an OpenRouter key Local models with Ollama

This exercise calls a model. Run it either way

Any model through OpenRouterOne key for Claude, GPT, Gemini, Llama and more. Pay per token, with a credit limit.A local open-source modelFree and private with Ollama. Weaker on hard tasks.

Which should I pick? →

The audit trail is a hook, not a trace. agent/hooks/audit.ts listens to three session events (call.settled, interaction.opened, interaction.settled), a pure function turns each into a minimised entry (identifiers and hashes, never document text), and a small store appends it to a per-session hash chain whose primary key is the event's stream position. That key is the decision that matters most: eve guarantees a position never changes, so a replayed event hits the key and is skipped, and the chain stays exactly as it was. Everything runs offline with eve's mockModel; the same project runs on your own OpenRouter key or a local Ollama model.

Answers Lesson 4: Compliance and auditability

Step-by-step answer

Stretch 90 min7 steps