Public hosting Cloud
The platform runs on managed cloud infrastructure. It is quick to start, grows with your usage, and needs no hardware on your side.
- Managed updates and monitoring
- Access from any approved device
- Capacity that scales with use
IOTA
AI software and services
We build AI software that runs in the cloud or on your own hardware: IOTA for Windows, IOTA Mobile, AI for games, custom model integration, enterprise AI platforms, and IOTA Orchestration. We also develop sector allocation strategies.
Point at a hub or a leaf.
What we do
What IOTA does
IOTA builds AI software and services for companies and individuals, and also develops sector allocation investment strategies. Each area below shares the same engineering base of data pipelines, model testing, and software that can run in the cloud or on local hardware.
Hub
Enterprise AI
Platforms for companies, hosted in the cloud or on infrastructure you control.
Enterprise AISkills
Slash-command add-ons, with the downloadable library at /skills.
Enterprise AIModels
Public, fine-tuned, or custom models connected to business systems.
Hub
Windows
A desktop AI app in two editions: Exe and PowerShell.
WindowsExe
The EXE edition is faster and can connect to cloud models.
WindowsPowerShell
The PowerShell edition is strictly offline.
01 / Enterprise AI Platforms
AI platforms for companies
We build specialized AI platforms that companies can purchase: assistants, document search, and workflow tools set up around a company's own data and processes. Each platform is available with two hosting models.
Private hosting On-premises
The platform runs on servers you control, in your own building or your own cloud account. Your data and models stay inside your network.
- Can run fully offline
- Fits strict data-handling rules
- You control the hardware and who has access
A platform can include:
- Chat assistants
- Search across company documents
- Report and document drafting
- Data extraction from files
- Connectors to internal tools
- Access controls and usage logs
- IOTA Orchestration retrieval
02 / IOTA for Windows
A desktop AI app for Windows
IOTA for Windows is built for ease of use. Everything is arranged as one core with branches for models, skills, connectors, knowledge, and memory. It runs from a USB kit and comes in two editions: a PowerShell edition that is strictly offline, and an EXE edition that can also connect to cloud models. IOTA for Windows is under continuous improvement, so features on this page may change between versions.
Offline from a USB kitPlug in the kit and run. No internet needed to chat.
Chat sidebar with projectsChats saved locally, grouped into projects.
Long contextLong documents and long chats fit in one conversation.
Built-in skills
See Skills
/chart /deck /pdf /ocr /transcribe /units /finSee Skills
Two editions
Both editions run from the USB kit. Pick what you need.
PowerShell edition
Driven by PowerShell for increased security and easier updates.
- Strictly offline. It does not connect to the internet or to online services.
- Local data only. RAG uses local databases only.
- Protects the model. Staying offline is a security measure that protects the active model on the USB drive.
EXE edition
Faster, with enhanced animations.
- Faster. Built for speed, with enhanced animations.
- Cloud models. It can connect to cloud models.
- One-click switch. Switch between local and cloud hosting with one click.
Parts of IOTA for Windows
Select any part of the tree to see what it is and how it works.
Interactive. Select a branch
IOTA core
What it is
The core is the app itself: the chat window, the model runner, and the settings that tie every branch together.
How it works
Each branch plugs into the core. Adding a connector, a skill, or a knowledge base gives IOTA a new ability without changing how you use the app. You keep typing in the same chat, and IOTA decides which branch a request needs.
Editions
What it is
IOTA for Windows comes in two editions: PowerShell and EXE.
How it works
The PowerShell edition is driven by PowerShell for increased security and easier updates. It is strictly offline and uses RAG with local databases only. This is a security measure that protects the active model on the USB drive.
The EXE edition is faster, with enhanced animations. It can connect to cloud models and switch between local and cloud hosting with one click.
Local models
What it is
The AI model runs on your own PC, not on a remote server.
How it works
IOTA loads open-weight models from the USB kit and runs them on your graphics card or your processor. You can switch models for speed or quality, and a CPU Only mode works on PCs without a supported graphics card. No internet connection is needed to chat, and your conversations stay on your computer.
In the EXE edition, you can also switch to a cloud-hosted model with one click and switch back to local the same way. The PowerShell edition runs local models only.
MCP connectors
What it is
MCP (Model Context Protocol) is an open standard that lets an AI app work with outside tools and data sources through small connector programs.
How it works
Each connector describes the actions it offers, such as reading a calendar, searching a drive, or querying a database. IOTA lists the connectors you have added, the model picks one when a request needs it, and IOTA shows you what it did. Connectors are optional, and you choose which ones to add.
Connectors that reach online services need the EXE edition. The PowerShell edition is strictly offline, so it does not reach online services.
Skills
What it is
Skills are focused add-ons for specific jobs. You start one with a slash command, or IOTA starts it when a request calls for it.
How it works
The skill does the task with tested code, for example drawing a chart or converting units exactly, and the model writes around the result. Charts, decks, and PDFs come out in the IOTA theme.
/chartBuild a chart from numbers or a table./deckMake a slide deck./pdfCreate a PDF document./ocrRead the text in an image or scan./transcribeTurn an audio recording into text./unitsConvert units exactly./finRun finance calculations.
Skills on this page is the overview. Open the Skills Library to search, filter, and download.
Knowledge bases
What it is
A knowledge base is a set of your own files, such as notes, manuals, or reports, that IOTA can search when it answers.
How it works
Add files to a knowledge base and IOTA indexes them on your PC. When you ask a question, IOTA searches all of your knowledge bases by meaning, pulls in the most relevant passages, and answers from them. Nothing is uploaded.
RAG pipeline
The Knowledge branch uses IOTA Orchestration, our proprietary RAG approach.
Memory
What it is
Memory holds notes and facts you want IOTA to use in later chats, such as your role, your projects, or your preferences.
How it works
Memory is a local collection of notes. You can add a useful chat to it, and IOTA retrieves from it in later conversations the same way it searches a knowledge base. It is stored only on your PC, and you can edit or delete it at any time.
Chats and projects
What it is
The chat sidebar keeps your conversations organized into projects.
How it works
Each chat is saved on your PC with its date and time. Group chats into projects, reopen any past conversation, and see how many chats each project holds.
Personalities
What it is
Personalities change how IOTA writes and lays out an answer.
How it works
Pick one to suit the task, such as a direct default, a researcher that gives findings and a confidence level, or a manager that lists action items. The facts stay the same; only the style and structure change.
Skills
Add-ons for specific jobs
Skills extend IOTA for Windows. You start one by typing its slash command, or IOTA starts it when a request calls for it. The skill does the task with tested code, and the model writes around the result.
Skill packages
A skill package is a zip of the skill files that ship with IOTA for Windows. Select a command to see what it does. These seven are ready to download:
Diagram, table, summarize, and files are in the library and are not ready to download yet.
Skill guides
A skill guide is a written instruction by IOTA - Joseph Bryant. You paste it into an IOTA chat or follow it yourself. A guide does not install anything.
The library groups guides by building apps, engineering, IOTA tools, and finance. Each card says whether that guide needs the internet. Most run offline.
03 / IOTA Mobile
Coming soon to iOSAn AI assistant that runs on your phone
IOTA Mobile is a free AI assistant for Android. There is no account, no subscription, no ads, and no fees. The model downloads to your phone once, and chat works offline after that. Your chats stay on the device.
Fast responses
Quick answers
Use the Tiny or Small model for everyday questions. Replies come faster and stay on the phone.
Deep Research
Slow, methodical reasoning
IOTA searches the web and reasons step by step on the phone's own processor, so an answer takes several minutes. Prompt and leave: you get a notification when it is ready.
- Private by default. Memory and saved chats are off until you turn them on, and they are stored only on your phone.
- Online only when you ask. Web search is off by default. When it is on, replies show their sources.
- Background running. Leave the app while a long answer runs, with a notification when it finishes.
- Models from Google and NVIDIA. Gemma and Nemotron models, eleven personalities, and a skill library.
IOTA Mobile is in testing on Google Play. The closed alpha is open to invited testers. Email us to be added. Requires a 64-bit Android phone running Android 9 or later. An iOS version is coming soon.
04 / AI for Games
Fine-tuned models for mobile games
We develop and fine-tune small models that run mobile games directly on the phone. Many consumer AI character apps are built around adult content. Our game models are built for safe-for-work play that suits a general audience.
- Small and on-device. Models are sized to run on phone processors, so play works offline with no per-play cost.
- Fine-tuned for the game. Each model is trained on the game's own characters, rules, and tone.
- Content guardrails. Game text is filtered for safe-for-work output, and players can report anything that slips through.
Closed beta
Iota Tales, our first title
A this-or-that space story game. You are the new captain of a colony ship, making one-tap decisions for the crew while keeping Fuel, Food, Happiness, and Currency in balance. An AI on your phone adds detail to each card and writes your captain's log.
New mode
Play as AI
In Play as AI mode, you are VESPER, the AI that runs the station, and you decide how to answer the crew's requests each day. After the first five days you choose a path: help the crew, quietly work against them while keeping their suspicion low, or openly fight them as they try to shut you down. Play as AI is in the current closed beta and unlocks after you reach day 20 as Captain.
05 / Model Integration and Custom Models
Connecting AI models to business systems
We connect AI models to the tools a business already uses, and we help choose the kind of model that fits the job.
Public models
Use an existing, publicly available model through its provider or as open weights. This is the fastest way to start.
Fine-tuned models
Start from a public model and train it further on your data, terms, and formats so it does better on your tasks.
Custom-trained models
Train a model built for one business case, when the task, the data, or privacy needs call for it.
- AssessDefine the task, data, and limits.
- SelectPick public, fine-tuned, or custom.
- IntegrateConnect it to your tools and data.
- EvaluateTest it on your real tasks.
- SupportMonitor, update, and retrain.
06 / IOTA Orchestration
Retrieval-augmented generation for your documents and data
IOTA Orchestration is our proprietary RAG approach. We set it up on your documents and data and support it in use. It finds the right passages before the model answers, so a smaller model can give grounded answers with less computing power.
Ingest. Files such as documents, notes, and reports are read and converted to plain text.
Chunk. The text is split into short passages, each about one idea, so each can be found on its own.
Embed. Each passage becomes a list of numbers called an embedding. Passages with similar meaning get similar numbers.
Retrieve. Your question is turned into numbers the same way, and the passages closest in meaning are found.
Ground. The chosen passages are given to the model with instructions to answer from them.
Answer. The model writes its reply from those passages, which keeps the answer tied to your sources.
IOTA expanded. IOTA Orchestration is our proprietary RAG approach. It searches all of your knowledge banks at once, lets you tune how many passages are used and how close a match must be, and runs on a single PC, your own servers, or the cloud.
Read my LinkedIn post on how RAG works →
Search across every source
Retrieval runs by meaning across all of your knowledge banks at once, not one folder at a time.
Tunable retrieval
Controls set how many passages are used and how close a match must be, so answers can favor precision or coverage.
PC, server, or cloud
IOTA Orchestration can run on a single PC, on your own servers, or in the cloud.
Corpus reading speed
How fast each setup reads a full knowledge corpus.
Tokens per secondLog scale
-
Pre IOTA Orchestration (Slowest)551 tok/s (~33,060 tok/min)
CPU-only embedding, original kit runner, 4 threads. 460 chunks / 69,415 tokens in 126.0 s.
-
Gemini Argon 4 (cloud, reading raw corpus)29,710 tok/min (~495 tok/s)
-
Gemini Flash 3.8 (cloud, reading raw corpus)65,720 tok/min (~1,095 tok/s)
-
Embedding speed (IOTA EXE default GPU embedding, 1 parallel slot)~9,700 tok/s (~584,000 tok/min)
460 chunks in ~7.1 s.
-
IOTA Orchestration (Embedding tuned)20,783 tok/s (~1,247,000 tok/min)
GPU embedding with 8 parallel slots. Full corpus: 744 chunks / 112,377 tokens in 5.41 s.
Tuned IOTA Orchestration reads about 42x faster than Gemini Argon 4, about 19x faster than Gemini Flash 3.8, about 2.1x faster than the default EXE embedding, and about 38x faster than the pre-orchestration CPU path.
Bar length uses a log scale of tokens per second, so the slower rows stay visible. NVIDIA RTX 3060 Ti 8 GB on a Windows desktop. Engine: llama.cpp b11800 CUDA 12. Embedder: embeddinggemma-2 Q8_0, 768-dim, chunks up to about 490 tokens. The tuned setting is 8 parallel embedding slots. Cloud speeds were measured in each provider's web interface reading the same corpus. Date: Oct 23, 2026.
Live speed race
Each row reads the same 112,377 token corpus at its measured speed.
Elapsed 0.0 sShown at real speed
-
Pre IOTA Orchestration (Slowest)551 tok/s (~33,060 tok/min)
-
Gemini Argon 4 (cloud, reading raw corpus)29,710 tok/min (~495 tok/s)
-
Gemini Flash 3.8 (cloud, reading raw corpus)65,720 tok/min (~1,095 tok/s)
-
Embedding speed (IOTA EXE default GPU embedding, 1 parallel slot)~9,700 tok/s (~584,000 tok/min)
-
IOTA Orchestration (Embedding tuned)20,783 tok/s (~1,247,000 tok/min)
The race is shown at real speed and stops at 10 seconds. Projected time is how long that row would take to read the full corpus. If motion is reduced, the 10 second result is shown right away.
Reading time calculator
Enter a document size to see how long each setup takes to read it.
Enter a size of zero or more.
One page is about 500 tokens. Bars use a log scale of the measured tokens per second.
- Pre IOTA Orchestration (Slowest)
- Gemini Argon 4 (cloud)
- Gemini Flash 3.8 (cloud)
- Embedding speed
- IOTA Orchestration (Embedding tuned)
Local rows were measured on an NVIDIA RTX 3060 Ti 8 GB. Cloud rows were measured in each provider's web interface.
Tutorials
Step-by-step guides
IOTA Mobile
Getting started
Install the closed alpha, pick a model, and send your first message.
Read the steps → IOTA MobileDeep Research
How prompt and leave works, with background running and the ready notification.
Read the steps → IOTA MobileMemory and saving chats
Turn on Save chats and Permanent memory, and manage what is kept.
Read the steps → IOTA for WindowsSetup
Run IOTA from the USB kit, choose a model, and add your files.
Read the steps →Downloads
Apps, skills, and early access.
Closed alpha on Google Play
IOTA Mobile
Free private AI for Android. Invited testers can join the closed alpha from this link.
Coming soon
IOTA for Windows
The desktop app and USB kit are not publicly available yet. Two editions: PowerShell (strictly offline) and EXE (local or cloud). Request early access by email.
Closed beta on Google Play
Iota Tales
Our first AI story game. Invited testers can join the closed beta from this link.
Version 1.2
Skills Library
Skill packages for IOTA for Windows, plus written skill guides. Search the library, then download a zip or open a guide.
Internal testing
The Apothecary
A game still in development. It is in Google Play internal testing. There is no public download yet.
About
Founded by Joseph Bryant
Founder, IOTA
Joseph Bryant is the Founder of IOTA. He works as a production engineer in aerospace, is a graduate of the University of Oklahoma, and is pursuing an MBA in finance.
- Founder, IOTA
- Production engineer, aerospace
- University of Oklahoma graduate
- Pursuing a finance MBA
Contact
Email or LinkedIn
For questions about enterprise AI platforms, model integration, IOTA Orchestration, our apps, or investment management, email us and we will reply.