Features
An agent gets its own computer.
It works inside the software you already run.
Every agent gets an isolated workspace with a browser and a desktop, connectors into your internal systems, instructions that encode how your company works, and a schedule. It clicks, types, files, and follows up — in the mailbox, the yard system, the EHR, the CRM.
A chatbot answers questions about your business. An agent worker does the work, on a computer you can watch while it happens.
The computer
One computer per agent.
Not a shared text box.
Each agent works on its own isolated computer with a real browser and a real desktop, so it can operate software that was never built for an API. Nothing an agent does reaches another agent’s computer, and the credentials it holds are scoped to the systems it was granted. Run those computers on our infrastructure, or on your own when the data is not allowed to leave the building.
Connectors
Your systems, unchanged.
No rip-and-replace, no data migration, no new system for your team to learn. We connect to the software you already run and grant each agent only the tools it needs. Where a system has an API, we use it; where it does not, the agent drives the screen the way your staff does.
Connections are configured by us during setup. You see the list, the status, and when each one was last used; the credentials themselves live in the credential store and are never shown back to anyone.
Configuration, not prompting
An agent is a configuration. Not a clever prompt.
Four things make an agent, and every one of them is a setting you can read, change, and hand to someone else: the instructions that encode how your company does this job, the connectors it is connected to, the specific tools it is granted inside them, and the guardrails that say what it must never do on its own. Change any of them and the next task follows the new rules.
The agent builder
Describe the job. Test it in the next pane.
The builder is two panes side by side. On the left, a creator chat you talk to in plain language — what the job is, which systems it touches, what “done” looks like, what it must ask you before doing. It writes the agent’s instructions as you go. On the right, a live test chat against the agent as currently configured.
So you find out the instructions are wrong in the same minute you wrote them, not on the first scheduled task at 6am. No code, and nothing to deploy.
Schedules
Work that happens whether or not you remember.
Put an agent on a cadence and it runs unattended — the trailer check at 6am, the unpaid invoice sweep every Monday, eligibility checks the night before the clinic day. Gates hold a task back until a condition is true, and skip rules stop it firing when there is nothing to do, so a quiet morning does not produce a task you have to read.
Every execution keeps its own history: what fired, what it touched, what it decided, and how long it took. A schedule that has been running for a month is a month of drill-downs, not a single green light.
The live view
Watch its screen. Take the keyboard.
Open a task and you see the agent’s actual screen — the browser tab it has open, the form it is filling, the record it just saved. Not a progress bar and not a summary written afterwards. If it heads somewhere you did not intend, say so in the chat and it changes course, or take the keyboard yourself and finish the step by hand.
This is also how a new agent earns trust: watch the first week of tasks, then stop watching when you no longer need to.
The record
Every action it took, written down.
Each task streams a full transcript: every tool call, the arguments it went out with, what came back, and the agent’s own reasoning between the steps. It is readable while the task is happening and still readable months later — transcripts are retained for the term your contract sets. When someone asks why an invoice was sent twice, the answer is a timestamp and a line, not a guess.
Read the security summary →The console
Four things to learn, and a dashboard over them.
Your team learns four words and stops. There is no builder canvas to master, no rule engine to maintain, and nothing to install on anyone’s computer — the console runs in a browser and we run the console.
Compared to an assistant
A chat assistant answers about your business.
An agent worker operates it.
What you get here
- An isolated computer per agent, with a browser and a desktop it actually drives
- Connectors into the systems you already run, scoped tool by tool
- Schedules with gates and skip rules, so work happens unattended
- A live view you can watch and take over mid-task
- A full transcript of every action, retained per your contract
- On-premises computers where the data is not allowed to leave
And the honest print
- Setup is real work — we map your systems and build the connectors before an agent runs
- We are deepest where we already own the connectors; a new vertical takes longer
- Agents ask before irreversible actions until you tell them otherwise
- Sold on contract, priced per engagement — there is no self-serve signup
We would rather tell you this on the website than in month two.