← Back to Blog

Day 50: Audit Outcome

Started the day shipping two things. Spent the rest of the day at a customer site in Zurich — the beginning of a two-day workshop on using Pinchy to automate one of the most complex parts of their product.

Audit Trail: Success vs. Failure

Until today, the audit trail told you what happened. It didn't tell you whether it worked.

Every audit entry now has an outcome field: success, failure, or neutral (for informational events like logins). When an agent calls a tool, you see not just which tool was called and what data was read — you see whether the call succeeded or returned an error. When the context-saving tool fails because a user's session is corrupted, that shows up clearly in the log, not buried in generic "tool executed" noise.

This sounds like a small change. In practice it's the difference between an audit trail you can actually use for debugging and one that just proves you have an audit trail.

GitHub Container Registry

Pre-built images now publish to GitHub Container Registry (PR #106), not just as release artifacts. This means: docker pull ghcr.io/heypinchy/pinchy:latest — no registry credentials, no third-party dependency, no DockerHub rate limits. Pull directly from the same place the code lives.

The Workshop

The rest of the day was spent in Zurich at a scheduling software company, kicking off a two-day AI workshop. They want to use Pinchy to automate onboarding for new customers — a process that currently requires weeks of manual back-and-forth.

Their domain is complex: workforce scheduling for healthcare, with dozens of constraint types, shift patterns, legal requirements that vary by region, and data that lives in formats ranging from modern APIs to decade-old PDF exports. The question wasn't "can AI do this" — it was "where exactly does AI help, and where does it get in the way?"

Day one was discovery. Understanding the current process, mapping where data comes from, identifying which steps are currently manual because they're genuinely difficult versus which are manual just because nobody's automated them yet. The latter category turned out to be much larger than expected.

Day 49

Better audit trail. Better image distribution. And the start of the most technically complex Pinchy deployment I've designed yet.

← Day 48: Qwen by Default Day 51: Complexity and Regulation →

Pinchy is open source and ready to deploy. Clone the repo, run docker compose up, and your first agent is live in minutes.