Automate the desktop work your team still does by hand.

Lapu works in the driver layer, not the mouse and keyboard. It talks to each application's native automation layer — Windows UI Automation, macOS AXUIElement — the same interface a screen reader uses. That layer exposes an app as named elements: the Save button, the Amount field, the third row of a table. Lapu addresses those elements directly instead of clicking pixel coordinates. So a moved window, a different resolution, or a new theme does not put an action in the wrong place, and a workflow does not drift between the first run and the fiftieth.

Two honest caveats. Lapu also supports driving real keyboard and mouse input, and there are apps where that is the right call rather than a degradation — SAP GUI is the clearest, because it means you never have to enable sapgui/user_scripting, a setting most security teams keep off. In that mode the agent occupies the session while it runs. Where an app publishes nothing to address at all — Citrix and VDI sessions, terminal emulators, some custom-drawn interfaces — reading falls back to OCR and vision, which is less robust.

And we publish no reliability benchmark, so treat this as a description of the mechanism, not a number. On the intro call we map your specific apps and tell you which ones land where.

Lapu AI ships as standard macOS and Windows installers and runs on each user's machine — physical or virtual. VMs are a supported deployment target, so bots can run on dedicated virtual machines. There is no server-side component to stand up; rollout works with your existing device-management tooling, and most teams are running within days, not quarters.

The Teams plan includes a shared usage pool, unlimited seats, a team management console, and centralized billing. Enterprise adds SSO/SAML, a custom SLA, and custom deployment. If you prefer the model you know from RPA platforms, per-bot billing is available too — you license each bot instead of per-user seats. Both are priced to your scope — that conversation starts on the intro call.

Usually alongside them, not instead of them. Those platforms are built for unattended, orchestrated volume — queues, retries, SLAs, central governance — and Lapu does not replace that. Where they get expensive is the long tail: the workflow one team runs twice a week, the legacy app nobody wrote a connector for, the process that changes every quarter. Each of those still needs a developer to build and maintain it and a bot to license, so the business case rarely clears. Lapu covers that tail at per-user pricing, with the person who already does the work describing it instead of filing a ticket. A rough rule: if your automation backlog is mostly long tail, start with Lapu; if it is mostly high-volume batch, keep the platform and use Lapu for what the platform keeps declining.

Files and workspace data stay local — there is no Lapu cloud storage. When a step needs AI reasoning, only the relevant context for that step is sent to the model endpoint. Every action is written to a local, exportable audit log. The full architecture is documented on our security page.

Yes. Sensitive actions — file writes, shell commands, desktop automation — are gated behind explicit approval prompts by default, and every action lands in the audit trail. The agent never acts beyond the scope it has been granted.

A two-week, fixed-scope engagement: 2–5 users, 2–3 of your workflows, hands-on setup help from us, and hours measured against your baseline. Pilot pricing depends on scope and is agreed on the intro call — no long-term commitment required.

Teams plans include priority support; Enterprise adds a dedicated account manager and a custom SLA. During the pilot you work directly with us on workflow setup.

Bring them to the intro call. We walk through the data-flow architecture, the permission model, and the audit trail with your IT or compliance stakeholders, and handle data-processing terms as part of procurement.