Day 13: The Thinking Day

Zero commits. No code. Just the questions that shape the next two weeks.

← Day 12: The Audit Trail Closes

Not every day needs a commit

Sunday. No code today. On purpose.

After two weeks of daily shipping — 270+ commits, 11 blog posts, 3 PRs merged — I needed a day to step back and think about direction. Not "what should I build today?" but "what should Pinchy become?"

Two features have been circling in my head all week. Both came directly from conversations with potential users.

RBAC: who gets to do what

Right now, Pinchy has two roles: admin and user. Admins see everything. Users see shared agents and their own personal agents. Simple. Works for a team of 3.

But here's what a CTO asked me on a call last week: "Can I give the marketing team access to a content agent without them seeing the engineering agents?" And the honest answer was: not yet.

Real RBAC means:

• Who can use which agents
• Who can configure which agents
• Who can see which data the agents access

This isn't just a permissions table. It's a fundamental question about how multi-tenant AI agents work. If Agent A has access to HR documents and Agent B has access to engineering docs, the permission boundary needs to be airtight. One leaked tool call and your salary data shows up in the wrong chat.

I spent most of the morning sketching data models for this. No code yet. The architecture needs to be right before the first migration.

Knowledge base indexing: the real enterprise feature

The second thing on my mind: what happens when a company has 10,000 documents they want their agents to know about?

Currently, Pinchy agents read files from their workspace directory. Upload a PDF, the agent can reference it. Works for 10 files. Doesn't work for a knowledge base.

What enterprise customers actually need:

• Point the agent at a folder, a SharePoint, a Confluence space
• Auto-index everything. Convert PDFs, Word docs, markdown, whatever
• When the agent answers, cite the source. "Based on HR Policy v2.3, section 4..."
• Keep it updated. New documents get indexed automatically

This is the feature that separates "cool developer tool" from "thing a company actually pays for." And it's the feature that multiple people have asked about in the first two weeks.

I'm not building it yet. But I'm designing it. The architecture decisions I make now — how documents are stored, how context is injected, how citations are tracked — will determine whether this scales or becomes a rewrite.

The value of not coding

Build in Public doesn't mean ship every day. It means be honest every day.

Today's honest update: I didn't write code. I wrote notes. I sketched database schemas on paper. I thought about what kind of product Pinchy should be in three months.

The best code I'll write this week will be better because of today.

What's next

Back to building tomorrow. The context refactor PR needs to merge. And then: starting the first prototype of agent-level permissions. Small scope, clean architecture, the foundation for everything that comes after.

Day 14: The Demo That Broke Everything →

Follow the build: github.com/heypinchy/pinchy