Instructions anyone can write
Write the steps like you'd explain them to a new hire. No flowcharts, no code.
EZAgents are AI teammates that follow instructions written in plain English, work in your apps and on the web, run on a schedule, and stop for a person's approval before anything important.
You choose the tools, the schedule and the approval points.

A practical AI teammate — with the instructions and boundaries your business needs.
Write the steps like you'd explain them to a new hire. No flowcharts, no code.
Run a task every morning, every Friday or on the first of the month.
Agents can work on any website — and you can watch or take over at any moment, in the same window.
Sending, paying, deleting — you pick the actions that need a person's OK, and who that person is.
Admins choose which AI models, apps and templates each team's agents can use.
Agents can run calculations and scripts in an isolated sandbox.
Coming soonAccount research and follow-up drafts are waiting before the day starts.
Match invoices to purchase orders and flag only the exceptions for review.
Draft responses from order history, with a team lead approving any refund.
Only as much as you allow. You decide which actions need approval — everything else runs, and all of it is logged.
No. Workflows are written in plain English. Your IT team may help set up the first app connections.
Yes. Your other software can talk to an agent the same way it talks to an AI model.
EZAgents runs on the Mastra agent framework. Shared browser sessions use a self-hosted Steel.dev browser; sandboxed code execution (powered by E2B) is in development. Agents can be used through the chat interface or called like a model through an OpenAI-compatible chat-completions API. Human-in-the-loop approvals attach to specific tool calls, and admins manage which models, MCP connections, templates and API keys are available.
Pick the problem that matters most today. Connect the rest when you're ready — everything is built to work together.