Your intelligence layer.
At home on your desktop.
Local models. Your own skills. Orchestration with built-in MCIR memory on connected theo requests. Keep the pieces together on your device, with cloud access when you connect your account.
A little help with the release.
Your model and a skill for the brief. Connected requests can also recall relevant MCIR memory; this offline preview does not access it.
A synthetic request is ready. No prompt or private data is collected.
Find your kind of intelligence.
4 examples to get you started
Explore all preview capabilitiesPublic theo capabilities and local setup examples, not a live account inventory. Cloud models are not automatically available to run locally.
Release desk
Draft a release brief
Synthetic notes attachedA model for the job.
Sample graph. Local execution depends on available models and tools.
Move, connect, and run pieces in the full workbenchYour project. A little company.
type ReleaseBrief = {
summary: string;
sources: string[];
openQuestions: string[];
};
// Keep the review with a person.
const status = "draft";Give the assistant a focused job.
Inspect changes. Run your tests. Keep the final say.
Explore cloud capabilities or prepare compatible local models through Ollama and Hugging Face. Inspect a model, review its requirements, and confirm installation in desktop.
01one place to work
Less switching. More shaping the system.
A conversation with your whole workspace.
Choose a model for the task. Bring your context and configured tools into a conversation without losing sight of where it runs.
explore the model choicesYour models. One familiar home.
Explore cloud capabilities or prepare compatible local models through Ollama and Hugging Face. Inspect a model, review its requirements, and confirm installation in desktop.
meet your model libraryA workspace for your routes.
Create and edit local orchestrators. Linked web orchestrators are read-only in desktop and need a connection.
try the interactive workbenchKeep the build close to the conversation.
Bring a configured assistant into the coding workspace. Inspect proposed changes and test real code before accepting it.
explore the developer workflow02local and connected, not interchangeable
Know where the work happens.
On your device.
Choose local models and build local skills around your own instructions and reference material. Create and edit local orchestrators for that workspace.
- Create, install, configure, and enable local skills.
- Edit local orchestrators; use them in supported chat and IDE flows.
- Local execution needs the relevant models and tools to be available.
Connected to your account.
Link your account for cloud services and access to web orchestrators. Connected theo requests use MCIR to recall relevant context within that owner scope, even as the selected model changes.
- Web orchestrators are read-only in desktop.
- Edit web configurations in the web app.
- Cloud services and web orchestrators require a connection.
03release availability
Choose the build for your device.
Use the approved installer for your architecture and read its release-specific setup guide.
macOS
Apple SiliconVersion 0.0.1-rc56 · Signed and notarized
macOS 11 Big Sur or later on Apple Silicon (M1 or newer).
Download for macOS.dmgSetup & release details (opens in a new tab)Windows
x64Public release pending
A verified installer and its system requirements will appear here when available.
Target platforms: macOS Apple Silicon and Windows x64. No Linux or Intel Mac build is offered here. Use the listed release requirements rather than assuming support for every system version.
04when a build is available
Start with one model and one useful skill.
- 01
Install the approved build
Check the architecture and system requirements on the release card. Follow that build’s setup instructions rather than an unrelated development-install guide. - 02
Choose local or connected work
Prepare a local model for on-device tasks, or link your account for cloud access. Check the dependencies of the tools you plan to use. - 03
Give the workspace your know-how
Create a small skill, attach the reference knowledge it needs, and review the permissions. Begin with synthetic material and a read-only task. - 04
Shape a route and inspect it
Edit a local orchestrator or inspect a linked web one. Keep the local/cloud distinction visible, and verify policy bindings in the environment that executes the request.
a few useful distinctions
A desktop workspace, not a different set of responsibilities.
Is the window above a live desktop session?
No. It is an interactive product preview with authored examples. Changing its tabs does not connect an account, install a skill, run a model, or save a configuration.
Can I edit my web orchestrators in desktop?
The desktop implementation distinguishes editable local orchestrators from read-only web orchestrators. Use the web app to change web configurations. Web orchestrators are unavailable offline.
Does everything work offline?
No. Offline use depends on the local model files and the tools the workflow needs. Cloud inference, web orchestrators, downloads, and network-backed tools require a connection.
Does memory belong to one model?
No. MCIR is theo’s built-in context layer, not a single model’s chat history. Connected requests can recall relevant facts, preferences, and decisions within the same authenticated owner scope. Local-only workflows do not imply access to cloud memory while offline.
Do gateway policies automatically apply to a local model?
No. Gateway Guardrails apply to the configured gateway path and its API-key binding. Running a model locally does not automatically put that request through the gateway.
Where can I learn to make my own skills?
Follow the skill development journey for visual composition, SDK authoring, testing, and publishing. The desktop Skills surface also supports local authoring and configuration.
Make the workspace yours.
Build a skill for a useful task, then choose the workspace that fits the way you work.