Nothing here asks to be taken on faith. Every entity carries evidence — a verbatim quote and the public page it appears on — so a person or an AI can check any line of the map against the source.
Thalatta
OrganizationAlso called: thalatta.ai
A small, senior AI studio in Cleveland, Ohio. It designs, builds, and maintains AI agents that do a real job inside a business, each backed by a memory layer of the company's own curated knowledge.
“We build AI agents that work inside your business — across your team and your whole customer base. Each one is focused on a real job and backed by a memory layer, so it knows how you actually operate.” — source
“Thalatta is an AI consulting shop based in Cleveland, Ohio that designs, builds, and maintains AI agents for real business work.” — source
Jeff Tirey
PersonFounder of Thalatta. He helped build the early internet and has spent 25 years making knowledge open, findable, and useful — from the early web, through digital attribution, into machine learning since 2010, and now AI agents.
“Today Thalatta is mostly one set of hands — mine, Jeff Tirey — with a trusted collaborator or two when a job calls for it.” — source
“I was working in machine learning — prediction, decision trees, the algorithmic guts of it — as far back as 2010” — source
Custom AI agents
ServiceAlso called: AI agent development · agents for business jobs
Agents built for one real business job each — from simple visible workflows, to managed platforms, to agents running inside the client's own cloud, to fully governed architectures for regulated data.
“We build AI agents in a range of shapes — and the memory layers behind them — from the simplest thing that works to the most controlled.” — source
“from the simplest visible workflow (n8n-style flowcharts) to agents on managed platforms (Anthropic, OpenAI, Gemini), to agents running inside the client's own cloud, to a fully governed two-room architecture for regulated data.” — source
Memory layers
ServiceAlso called: memory layer · knowledge base · the sea, charted
The knowledge behind the agent: the company's own hard-won context, curated so an agent operates like someone who knows the business. Built as OKF wikis, vector databases, or knowledge graphs, chosen to fit how the knowledge is shaped.
“A memory layer is what a good employee keeps: your playbook, your customers, what they worked out yesterday.” — source
“A memory layer runs through every agent above — it's the durable knowledge the agent works from.” — source
The trust scaffold
ConceptAlso called: permissions, source trail, and human threshold
The governance around every agent: written permissions for what it can and can't touch, a source trail behind every answer, and a defined point where the agent stops and asks a person.
“The agent is the worker; the scaffold is the workbench around it.” — source
“written permissions for what it can and can't touch, a source trail behind every answer, and a defined point where it stops and asks a person” — source
Maintenance and monitoring
ServiceAlso called: agent maintenance retainer
Ongoing measurement and tuning of shipped agents — what the agent reads, what it touches, whether its answers hold up, and what each run costs — as a monthly retainer after a fixed-scope build.
“We watch what the agent reads, what it's allowed to touch, and whether its answers still hold up — and we adjust before drift turns into cost.” — source
“Discovery and build are fixed-scope; keeping it alive is a monthly retainer.” — source
The glass agent
ConceptAlso called: transparent AI agent · customer-facing agent you can see through
Thalatta's signature architecture: a customer-facing agent that works only from an approved packet (the caller's own record plus a curated slice of the memory layer), holds no keys, sources every answer, and acts within written permissions.
“Before it answers, a trusted broker assembles a small packet — the caller's own record, plus the right slice of the memory layer for the job — and hands it in through code.” — source
“It holds no customer list, no database connection, and no keys. The server hands it your record, and only yours, before it answers.” — source
Broker + blind agent
MethodAlso called: two rooms · control-plane / data-plane separation · trusted-broker / untrusted-agent clean room
The separation behind the glass agent: a trusted broker holds the keys and assembles an approved packet; the agent runs in an isolated sandbox with no keys, no database, and no path out. Context, not access.
“The agent runs blind, in an isolated sandbox — no keys, no database, no path out. A trusted broker in the next room holds the credentials, decides what the agent is allowed to know, and keeps the log.” — source
“a trusted broker holds the keys, the database, and the business logic; the blind agent runs in a sandbox — an isolated box with no keys, no database, and no path out — and only ever receives an approved packet and returns a result.” — source
Open Knowledge Format
ConceptAlso called: OKF · LLM wiki · knowledge base as files
An open, vendor-neutral standard from Google Cloud (v0.1, June 2026) for curated knowledge as plain markdown with YAML frontmatter — readable by humans and agents alike. The format Thalatta's memory-layer wikis and these islands are written in.
“The flavor we prefer (it follows Google's Open Knowledge Format); this site's own knowledge runs on one.” — source
“OKF is an open standard for the curated knowledge AI systems need, published by Google Cloud's Data Cloud team (v0.1, June 2026).” — source
The glass demo
ServiceAlso called: Hýalos · inside-out marketing · the glass campaign
A live marketing campaign run in the open: a prospect signs in, sees the exact record Thalatta holds on them, asks a glass agent anything, and sets their own status. Every change is audited.
“We call it inside-out marketing: the record we keep on you is yours — to read, to question, to change.” — source
“This website is one of our agents.” — source
Private and local models
ConceptAlso called: privately hosted models · local LLMs · inference inside your walls
The private end of the model choice: a privately hosted model or a local open-weight LLM on the client's own hardware, reached for on cost at high volume or when regulation means data can't leave the network.
“But the model can be private too — a privately hosted model, or a local open-weight LLM on your own hardware.” — source
“And if your data can't leave your network, that answer can include privately hosted models or local LLMs — inference that stays entirely inside your walls.” — source
The engagement path
MethodAlso called: the work · learn the need → set the targets → build → run → stay
How an engagement runs: learn the one job worth doing on-site, set targets before any code (accuracy, turnaround, cost per run), build the agent and its scaffold, run it with logs and an audit trail, then stay and keep it sharp.
“We come in, learn the one job worth doing, set the targets it has to hit, and build the agent and the scaffold around it. Then we stay — measuring results, tuning for accuracy, and keeping it cost-efficient as your business and the models keep evolving.” — source
Agents already shipped
StoryAlso called: RFP agents · HR agents · event agents · interview agents · SOP agents · ad-approval agents
Proof of practice: shipped agents that draft RFP responses from approved language (and have won new business), answer routine HR questions, move knowledge through conferences, train sales teams with realistic practice, apply company standards before work ships, and watch brand safety on creative.
“Draft first-pass proposal responses from approved language — and have won new business.” — source
“RFP agents that draft proposal responses from approved language (and have won new business) · HR agents that answer routine questions · event agents that move knowledge through a conference · interview agents that train sales teams with realistic practice · SOP agents that apply company standards before work goes out · ad-approval agents that watch brand safety on creative.” — source
The 20-minute conversation
ServiceAlso called: book 20 minutes · coffee · the first step
The contact channel and the first step of every engagement: a 20-minute conversation about how a specific job gets done today, where it's slow or error-prone, and what data it touches — ending in a straight answer on whether an agent is worth building.
“Not a pitch deck — a real conversation. Tell us how the work happens in your business and we'll tell you, honestly, whether an agent fits and where we'd start. If it wouldn't help, we'll say that too.” — source
Cleveland, Ohio
PlaceThalatta's home base. On-site work with client teams across Greater Cleveland.
“A small AI shop in Cleveland, Ohio — and not new to this.” — source
“our home base. We work on-site with client teams across Greater Cleveland and Northeast Ohio (Akron, Canton, and the surrounding region).” — source
Northeast Ohio
PlaceAlso called: Greater Cleveland · Akron · Canton
The region Thalatta serves on-site: Greater Cleveland, Akron, Canton, and the surrounding area, with Ohio and the Midwest within driving distance and the rest of the US remote.
“We work on-site with client teams across Greater Cleveland and Northeast Ohio (Akron, Canton, and the surrounding region).” — source
"Thalatta! Thalatta!" — the Ten Thousand reach the sea
StoryAlso called: Θάλαττα! Θάλαττα! · The sea! The sea! · the Anabasis
Xenophon's Anabasis, 401 BC: ten thousand Greek soldiers fought their way home from Persia, and from Mount Theches the vanguard saw the Black Sea and shouted the word that meant the way home. Thalatta is Attic Greek for 'the sea' — for a business, the memory layer is the sea: where hard-won knowledge accumulates and nothing of value is lost.
“Thalatta is the Greek word for the sea. A company's memory should work like one.” — source
“From Mount Theches the vanguard saw the sea and shouted "Θάλαττα! Θάλαττα!" — "The sea! The sea!" — meaning the way home.” — source
