---
title: What Thalatta Does
summary: Thalatta is a small, senior AI studio in Cleveland, Ohio. It builds AI agents that do a real job, backs each with a memory layer so it remembers what the business knows, and works on-site to keep them earning their keep as the business and the models change.
tags: [thalatta, what-we-do, services, memory-layer, glass-agent, on-site]
updated: 2026-06-30
source: founder direction 2026-06-30 (gold overview), content/homepage.md, content/services.md, content/00-guiding-light.md, architectures.md, glass-agent.md
---

# What Thalatta does

**In one line:** Thalatta builds AI agents that do a real job, remember what your business
knows, and keep earning their keep — built with you, on-site, and tended over time.

Thalatta is a small, **senior AI studio based in Cleveland, Ohio** — the practice of its
founder, **Jeff Tirey**. It builds AI agents that do real work inside a business — across a
team and an entire customer base — with each agent focused on a **real job** rather than acting
as a general-purpose chatbot. It works with small and mid-sized businesses, **on-site and
contract-first**, and stays on to maintain what it builds. The posture throughout:
**context, not access — appropriate access, not maximum access.**

## The memory layer

Every agent is backed by a **memory layer**: the company's own knowledge — its playbook, its
customers, its hard-won context — handed to the agent so it's *given* what it needs instead of
going fishing for it.

The vivid version: **raw AI is a brilliant new hire who forgets everything overnight; the memory
layer is what a good employee keeps.** (For the format and method behind it — assembled up front,
not retrieved reactively — see [the memory layer / OKF](about-okf.md).)

## The full range — from simple workflow to fully governed system

Thalatta builds across a full range, from the **simplest workflow you can watch** end to end to a
**fully governed system** — including the **[glass agent](glass-agent.md)**: a *"two rooms"*
design where a trusted broker hands the agent only an approved packet, and the agent **holds no
keys and no database**. The principle is constant at every size: **context, not access.**
(The forms this can take are laid out in [the architecture](architectures.md).)

## Building is only half of it

Thalatta comes **on-site and works alongside your team**. It sets **measurable targets up front**
— accuracy, turnaround, cost per run — and then it **stays**. Agents drift as your business
changes and as the models improve, and **tokens aren't free**, so the work is maintenance, not
"launch and walk away." Thalatta tunes the agent and its scaffold over time so it keeps earning
its keep.

None of it is fluff: **every agent has a cost and a value, and both are kept in plain view.**

## Proof — agents already shipped

Thalatta has shipped agents that:

- **win RFPs**,
- **answer routine HR questions**,
- **move knowledge through conferences**,
- **train sales teams**,
- **enforce SOPs before work ships**, and
- **watch brand safety on ad approvals**.

And the proof is live: **thalatta.ai itself runs on a glass agent you can sign in and talk to.**

## What a client can hold Thalatta to

- **Wins you can see.** Every answer ties back to a source — no black box. You can watch this
  in [the glass demo](the-glass-demo.md).
- **No runaway autonomy.** The agent works inside boundaries you set; a person owns the line.
  (Never "fully autonomous.")
- **Targets in plain view.** Accuracy, turnaround, and cost per run are set up front and kept
  visible — every agent's cost and value stay legible.
- **Built to be maintained.** A system your team can run, watch, and grow — not a demo that
  rots in a quarter.

The first step is always a 20-minute conversation (see `/coffee`).