Quickstart
Get your first agent running in a few minutes.
API
Build and drive agents from Claude Code, Cursor or your own scripts.
How it works
1
Pick a job, or describe one
Install a template from the library, or start from scratch and let the agent configure itself from your description.
2
Connect your tools
Gmail, Slack, Notion, Stripe and hundreds more, one click each. Or any MCP server, or any API.
3
It runs while you sleep
On a cron, on a webhook, or whenever you message it. It pauses to ask you when something needs your call.
The model
Three words carry most of the product. A workspace is your account. It holds your plan, your members, your Boxes and your agents, and it runs on one private computer in the cloud, isolated from every other account, that every agent inside it shares. You can belong to several workspaces, personal or shared with your team, and switch between them from the pill in the top left corner of the app. An API key is scoped to exactly one workspace. A Box is a room of your workspace: the rectangle your agents sit on, on the board. It groups them by team, by client or by project. A Box is instant and free, so you make as many as the work needs. It is a way to arrange agents, not a wall between them: every agent of the workspace runs on the same machine and shares the same folder and the same database, whichever Box it sits in. An agent is an AI worker inside a Box. It has a name, a system prompt, a model, its own memory, its own skills, its own files and its own database. It talks to you in a chat, and it can call the other agents of the workspace by their handle. Everything else hangs off those two: skills it loads on demand, apps to your tools, scheduled tasks and webhooks that make it run without you, and templates that package a working setup for someone else to install.Core concepts
Every term in one page.
What makes it different
Any tool on any model
Bring your own API key, your Claude or ChatGPT subscription, or use the credit included in your plan. Swap the model without rewriting the agent.
You approve what matters
An agent can pause mid-run to ask a question or request an approval, even at 3am on a scheduled run. It resumes from your answer.
Memory that compounds
Agents write down what they learn and recall it later. Correct one once and it stays corrected.
Secrets stay on your machine
Credentials live on your own computer in the cloud, not in Rerun’s database.