Building toward agentic maintenance.

Amphidata is an AI-native, fully agentic maintenance platform. Headless at its core, interface-rich at its surface. Managers and technicians interact with it through natural multimodal interfaces (voice, video, images, text), while autonomous agents handle the work that has historically sat on humans: scheduling interventions, tracking asset status, generating work orders from observed conditions, guiding checklists in real time, surfacing alerts from telemetry, and managing inventory consumption without anyone needing to update a database.

The end state is a platform where technicians walk up to a machine, talk to Amphidata about what they see, and the system handles everything else. The manager sees the resulting work, not the bureaucratic plumbing around it.

Agentic platforms in industrial software will not be sold to the market in their final form. They will be built into it.

Industrial maintenance is a conservative buying environment. Plant managers evaluating a CMMS from an early-stage company are already taking a leap. Asking them to simultaneously accept a paradigm shift in how their teams operate is a compounding risk most procurement processes cannot absorb. A headless, voice-first, agent-driven platform on day one is a thesis the market is not yet ready to underwrite, no matter how technically defensible it is.

So we start with something the market already understands: a well-built, modern CMMS with a polished web interface. Work orders, asset hierarchies, checklists, maintenance schedules, team management, calendar views. The kind of system a maintenance manager can benchmark against the incumbents and feel oriented within minutes.

On top of that substrate, we layer AI. Not as a feature, as a foundation. Voice assistance for technicians on the floor. Automated checklist generation from manufacturer documentation. Pipelines that turn field observations into structured work. Agents that watch for patterns in asset behaviour and propose interventions before anyone has to ask.

Rome was not built in a day. Neither is an agentic platform that industrial buyers trust.

Three reasons.

Adoption. The buyer needs something they can compare. A modern CMMS is a known quantity. Voice and AI as embedded capabilities are upside. Headless agents managing a plant is a leap. We are selling the first, demonstrating the second, building toward the third.

Training substrate. Every work order created, every checklist completed, every asset status change, every photograph attached to a maintenance log is the behavioural data the agentic layer needs to operate safely. A conventional CMMS, well built, is the training environment for the agents that will eventually run on top of it. The interface is not a phase we will discard. It is the data generator that makes the next phase possible.

Trust. Autonomous agents in an industrial setting are not a UX problem, they are a safety and liability problem. A manager will not delegate the scheduling of a turbine inspection to a software agent until they have spent months watching that agent surface correct recommendations inside a system they control. The web interface is the period during which agents earn that trust, in public, with every interaction visible and correctable.

The two phases of Amphidata are not separate products. They are the same product at different points in its lifecycle.

Phase one, the CMMS with AI assistance, captures the structured behaviour that makes phase two possible. Every interaction is implicit ground truth: which alerts a technician acts on, which suggestions a manager overrides, how a checklist actually gets filled in the field versus how the manual says it should be. This is the signal that a generic foundation model cannot get from the open internet. It can only come from real plants, real shifts, real machines.

Phase two, the agentic platform, takes that signal and starts to act on it. Not by replacing the manager, but by removing the administrative burden that prevents the manager from doing the work they actually care about. Less time in a database. More time on the floor.

A common pattern in enterprise software is to ship a product, then add AI as a paid tier on top. We reject that pattern.

We believe AI is a basic right of the platform, not a luxury feature gated by pricing. Every technician, on every tier, with every work order, has access to the same intelligence layer. This is not a marketing position. It is a structural decision that shapes how we price (per plant, not per user), how we build (agentic primitives at the core, not bolted on), and how we sell (voice-enabled, agentic, multimodal, no upsell wall between the user and the intelligence they need to do their job).

If AI were optional, we would be selling a CMMS with an AI module. We are not. We are building a maintenance platform whose substrate is intelligence, currently expressed through a familiar web interface, eventually expressed through whatever interface the user prefers, including no interface at all.

The base CMMS is largely built. Asset categories with AI-extracted checklists, scan-to-voice pipelines, team management, work order assignment with calendar view, a technician app. The voice assistant module is functional. Maintenance schedules are also implemented. Inventory integration and shallow IoT are queued as the next substrate expansions, because the agentic layer needs spare parts and telemetry as first-class objects.

We are not betting on agentic AI as a feature. We are betting on it as the shape of the category. And we are building the only thing that survives in that future: a system of record that earns its place by being useful on day one and indispensable by year five.