Fillory · engineering

Agents for the job your tools said they'd do.

Fillory is an agentic engineering firm. We design, build, and govern custom agents for work that off-the-shelf software claims to handle and doesn't.

You already bought the tool. It missed things, or it flagged everything, or it needed someone watching it full time. We build the agent that actually does the job, scoped to your domain, governed so it stays accurate, and running where you need it.

What this looks like

A property management firm came to us with a fraud problem. Scammers copy their rental listings onto Facebook Marketplace and Craigslist, collect a deposit and first month's rent from someone who believes they're renting a real house, and disappear. It's gone as far as a family signing a lease and moving into a property the scammer never owned.

Products exist that claim to catch this. The firm bought them. They didn't work.

We built an agent that watches Marketplace and Craigslist for their properties, tells a fraudulent post apart from a legitimate re-list, and texts them the moment it finds one.

The hard part was never the scanning. It's the judgment. An agent that hallucinates flags real listings and burns the firm's credibility. An agent that drifts quietly stops catching the pattern it was built for. That's what governance is for, and it's in every agent we ship.

The laboratory

We don't write you a script and wish you luck.

Fillory agents are built inside the Fillory Agentic Operating System — our own infrastructure for building, governing, and running agents. It's the same system our own products run on.

Persistent memory. It remembers the things you care about, the difference you make.

Drift detection. When the agent's behavior moves, we see it.

Truth scoring. An agent that starts making things up gets caught before you do.

Hash-chained audit trail. Every decision on the record, in order, tamper-evident.

Off-the-shelf tools give you a black box and a support ticket. Fillory hands you a system that can show its work.

How an engagement works

Fifteen minutes. What you already run, what's breaking, what must not exist.

A written scope. What we build, what we won't, what you own afterward.

We build it. You see it work against real conditions before it goes live.

Who this is for

You have a job that matters, you bought the software that was supposed to handle it, and it didn't. Or you're about to put agents into production and you need them to stay accurate when nobody's watching.

Not for: a cheaper version of software that's already working for you.

Talk about a build

Fifteen minutes. What you already run, what's breaking, what must not exist.