Make Your Agents Brilliant
If you're building multimodal agents that can navigate GUIs, do tasks on the web or a computer then you need AgentTutor. Spin up a containerized desktop, show agents how to use any site or app and get the data you need to train genius level agents.
Join the Beta for FREE Now!
AgentTutor makes it easy to spin up a Linux desktop and teach your agent to use SaaS apps and desktop apps.
Our annotation model gives you automatic annotations. You can edit or change those annotations with ease. Approve actions or hide them if there was a mistake or use the data to teach the agent how to avoid mistakes.
Get actions like clicks, double clicks, drag and drop with precise click data, along with text typed. Coordinates, annotations, screenshots, all in one data pack that's exportable to many formats.
We hide the data in BOTH the screenshot and the text data. Designate anything you want as a secret. The model can learn to fetch secrets instead of training on PII, so you data is safe from prying eyes or accidental leaks from the model.
While today's agents have gotten smarter and smarter, there's just no way that it can know every website. If you have a corporate app that the model has never seen, you have to train your agent get good at that app, so you can build agentic workflows inside your enterprise. AgentTutor makes it possible.
Give the agent a task you want to teach it and then show the agent how to do it. Record all the data and then use it to fine tune an agent to precise actions in any app.
Hide Personally Indentifying Data
Get automatic annotations on all actions
Sometimes you make a mistake. You can easily hide that action from the agent
Export data in multiple formats from OpenAI, to ms-swift, and many more
We can also launch an agent to learn automatically. The agent does the task and you simply correct it. After an approved agent run, the system kicks off a small training job and moves the agent further along in its learning journey. No demonstration needed other than the agent doing the work and you correcting it.
Let them do the learning automatically
After the task, go in and correct their mistakes
Register your agent and fine tune it automatically
Fine tune the best multimodal models fast with our Rust based backend
