Your ML workspace is missing a tool. Describe it, and it is yours to keep.
Kernels, models, datasets and notebooks, in one workspace, native on macOS, Windows and Linux.
Local-first, and specific about it: your files, datasets and local models stay on your machine. Models run where you choose: local GGUF or ONNX, Hugging Face or hosted. No hosted service is used unless you connect one.
Software is changing shape
Classic tools ship a fixed set of features and you wait for the roadmap. IVYX flips it: the platform gives you a governed core of data, models, compute and workflow, and you extend it with AI to fit exactly what you are doing. The advantage is no longer how many features you ship; it is having a safe place to build your own.
How it works
A short loop you repeat, not a setup you fight.
Work in the workspace
Open datasets, notebooks and models. Query files with SQL and run code on managed kernels. The day-to-day is already here.
Need something? Build it with AI
Ask the built-in assistant to create the node or extension you are missing. It scaffolds a real, runnable capability, not a throwaway snippet.
You approve, then it runs
Every capability runs through policy, approval and audit. AI-built features cannot touch your data or the core without passing the same gates as everything else.
It is yours, and reusable
Keep it, version it, publish it to the marketplace. Your workspace grows with each thing you build.
Nothing the AI builds runs until you allow it
The reason AI-built features are safe to keep and share.
Every ivy-node, every extension and every AI-built feature reaches your data through one gate: the capability pipeline. Open and extensible on top, governed underneath.
Policy
Each capability declares its risk; the gate enforces what is allowed to run.
Your go-ahead
High-risk actions pause for a human before they run, not after the damage.
Audit
Every call is traced end to end: who ran what, when, with which result.
Isolation
AI-built work only reaches the data and capabilities it was granted.
One gate, on a machine that is not a platform. That is why the evidence for an ML model and the evidence for an agent run come out of the same call boundary.
MLOps
The model, under policy
Do the ML work on your own machine, and still pass the audit.
Every dataset a run read, every gate decision and every approval is already on one record, because every call goes through the same gate. Nothing to instrument, and nothing to move onto a platform first.
Data you can name
A run's record names the datasets it read and the artifacts it wrote, each with its content digest.
Gates that carry their number
A gate does not only fail. It records the number it was decided on and the bound that number was compared against.
From a model version to its evidence
Ask for a promoted model version and get one signed package covering the run that trained it and every gate that judged it.
It says what it cannot answer
A question the record cannot answer comes back as unanswered with its reason, never as a blank.
LLMOps
The agent, under policy
The policy is written with the agent and ships to production with it.
A network proxy cannot see an agent that has not left the developer's machine yet. The moment it is being written is the moment the policy is decided, so that is where the gate belongs.
The gate is not optional
Every tool call passes it, from the agent and from every panel. The path that skipped it was deleted, not switched off.
Policy that travels
The same rules export as gateway policy: the file you enforced on your machine is the file production enforces, unedited.
A run reads back as a tree
Every governed call of a run, nested under the turn that asked for it, with the refusals in place.
Local models, with a number
Score a prompt against your own golden set on a local model: deterministic, offline, and no token bill.
See it in action
Three building blocks that turn a workspace into a platform you extend. Each one ships as an extension, so everything you install and everything you build with AI passes the same gate.
ivy-node
AI-powered atomic components.
ivy-node gives you flexible, reusable building blocks of code, generated with AI, with manifest validation and typed capabilities.
- Generate AI-powered atomic components with manifest validation
- Discover, search and register nodes in a centralized registry
- Test and validate nodes before deployment
- Version, reuse and try components from the marketplace
ivy-agent
Bring workflows to life.
Turn code components into intelligent workflows. Design, test and run them, with governance built in. From simple background tasks to complex operations, full control is in your hands.
- Transform ivy-nodes into runnable workflows
- Approval workflow and policy-based risk assessment
- Artifact store, observability and audit trails
ivy-mcp
Add MCP power to your project.
ivy-mcp simplifies how your projects connect with other systems. Write your own MCP methods, publish your MCP server, and manage the tools it exposes.
- Develop custom MCP methods
- Publish and serve your MCP server
- Connect with external systems like Claude Desktop and Claude Code
Why people choose ivyx
Built around one idea: a workspace that grows with you.
Three marketplaces, one workspace
Counted from the ivyx registry at the last build. Everything here installs into IVYX Studio without leaving it.
Extensions
Data, models, notebooks and dev tools you can add, or build your own with AI.
Browse29Ivy Nodes
Reusable compute nodes, searchable by capability and ready to drop into a workflow.
Browse107Education
Guided, hands-on courses: one notebook per lesson, with a tutor that asks you to predict before you run.
BrowseWays to bypass the gate. Every tool call passes it, from the assistant and from every panel; the path that skipped it was deleted, not switched off.
Native desktop app for macOS, Windows and Linux, with the same workspace on each.
Local-first: your data and models can stay on your machine.
Built for how you work
Whatever your role, ivyx meets you where you are, and extends to what you need next.
Data Scientists
Explore datasets, prototype in notebooks and query files with SQL, without standing up infrastructure.
ML Engineers
Run on managed kernels, swap local and hosted models, and manage Docker containers and images from one place.
Researchers & Students
A free, local-first workspace to learn, experiment and share your work.
Two ways in
Use IVYX Studio yourself, or build your own product on the same governed core.
Use IVYX Studio
The desktop workspace for data, models and notebooks, free while we are in beta. Extend it with AI as you go.
Download for freeFree while we are in beta, and no credit card. Downloading needs no account.
Build your own product on it
Give your own product a governed core your customers can safely extend with AI.
Talk to usFrequently asked
How is this different from Jupyter or a normal IDE?
IVYX Studio does the notebook work you would do in Jupyter or JupyterLab: kernels, notebooks, datasets and SQL over your files. Around that sits a full dev toolchain with an editor, Git, a terminal and a debugger. The difference is what happens when you hit a gap. Instead of waiting for an extension to exist, you describe the capability you need and the assistant builds it as a real ivy-node. It runs only after it passes the gate, and then it is yours to keep, version and publish.
How is an AI-built feature kept safe?
Every capability passes the same gate: policy decides what is allowed, high-risk actions ask for approval, and every call is audited and isolated from the core.
What does ivyx do for MLOps?
It gives a local run the record an audit asks for. Because every call goes through the same gate, a run already names the datasets it read and the artifacts it wrote with their content digests, and each gate keeps the number it was decided on next to the bound it was compared against. Ask for a promoted model version and you get one signed package covering the run that trained it and every gate that judged it.
What does ivyx do for LLMOps?
The policy is written with the agent, not bolted on later: the gate sits at the call boundary on your machine, so it sees an agent that has never left it. The same rules export as gateway policy for production, a run reads back as a tree of governed calls with the refusals in place, and you can score prompts against your own golden set on a local model, offline and deterministic, with no token bill.
What is the Education marketplace?
Guided, hands-on courses you take inside IVYX Studio: one notebook per lesson, with a tutor that asks you to predict before you run. Browse the library on the site, then open the app and download a course from the Education marketplace. Courses are free, like the rest of the beta.
Is IVYX Studio free?
Yes. ivyx is in public beta: the desktop app and the full extension marketplace are free while we are in beta, and the editor itself stays free afterwards. Pricing for hosted AI and team governance will be announced before general availability.
Which operating systems are supported?
Native desktop builds for macOS, Windows and Linux.
Can I use it without installing?
A browser version is coming soon. It runs IVYX Studio right in your browser with nothing to install. The desktop app stays the full experience, with native terminals, local files, git and offline use.
Do I need an account?
Downloading needs no account. IVYX Studio asks you to sign in when you first open it, and during beta that account is free to create. It is how the app licenses itself on each machine.
Where does my data live?
IVYX Studio is local-first: your files, datasets and local models stay on your machine unless you connect a hosted service.