Software should use software.
Software has spent decades asking humans to adapt.
Every application has its own interface, rules, menus, permissions and workflows. People have become the integration layer between systems that were never designed to work together. That model is reaching its limits.
UI is passé.
The next generation of software should operate existing software on behalf of the people using it.
But substituting unpredictable, expensive agent calls for human clicks is not a complete solution. A model that clicks for you still pays to think, every time, and may not click the same way twice.
Execution needs infrastructure.
Reasoning should produce reusable systems.
Language models are valuable because they can interpret unfamiliar environments and solve novel problems. Repeated execution is a different problem. A model should not have to solve the same interface sequence indefinitely.
Gravitas turns each workflow a model discovers into a map that runs the same way every time. Reasoning models handle discovery, uncertainty and exceptions. Validated maps handle repeatable operations. Every successful reasoning run becomes a draft for a person to review, and the map grows with use: self-improving, human-sanctioned.
Use intelligence to build the process. Use determinism to run it.
The first time. A reasoning model explores the interface, finding its way through dead ends. It is slow and paid for call by call.
Up to 8 model callseach one paid for
Reviewed. A person checks the run and approves it as a map.
Every time after. The reviewed map runs the same steps directly, in seconds, with zero AI calls.
0 AI calls20% faster. 100% cheaper.
The maze costs everyone.
Gravitas began with a problem its founders experienced firsthand. University systems are fragmented. Ordinary tasks require students and staff to act as messengers between platforms, departments and processes. So we began building the software we wanted to use.
CampusWay is the first product built on a larger technical and commercial thesis: agentic automation becomes practical when learned workflows run without paying for model reasoning every time. It works inside the university portals students already use, read-only and draft-only, and it never presses the final button for them.
- 97
- campus workflows, recognised from thousands of phrasings.
- 0
- AI calls on a mapped run, which finishes in about 3 seconds.
Internal benchmarks, October 2026.
Three models carry the thesis.
Small models of our own, so your workflows stop paying for a large one.
CampusWay proves the idea; three models are how it scales. Each one takes over a job that would otherwise need a large reasoning model or a person, and does it with something smaller, cheaper and easier to audit. Argus is trained and benchmarked; Talos and Nomos are in training.
- ArgusKnows which task you mean.
Recognises which campus workflow a request means, across thousands of phrasings, or turns it away, without calling a reasoning model. In benchmarks, every held-out, human-written request it routed went to the right workflow, from a model small enough to run in the browser.
Trained and benchmarked
- TalosOversees every run.
Watches each run as it happens and answers with one of a small, fixed set of commands: continue, pause, stop, retry an alternative, escalate, or switch a failing workflow off for everyone at once.
In training
- NomosDraws the map.
Turns successful reasoning runs into draft workflows, proposes new phrasings and repairs, and charts each campus. It proposes; people approve.
In training
The people building Gravitas.
Four co-founders who met the university maze firsthand, and set out to build the way through it. Between us: the engineering, the operations and the finance to take it from the first campuses to many.
Nikolas Routh
Chief Executive OfficerCo-founder
Griffin Adelmann
Chief Technology OfficerCo-founder
Lorenzo Camilli
Chief Operating OfficerCo-founder
Benjamin Bo
Chief Financial OfficerCo-founder
Beyond higher education.
Universities are the starting point. Healthcare systems and financial institutions also depend on complex software environments, repeatable procedures and sensitive information. The same deterministic execution layer extends into these markets through sector-specific implementations.
Higher education.
Now
CampusWay, inside the portals students already use: read-only, draft-only and fully logged.
Healthcare.
Later
Clinical and administrative workflows, within the safety boundaries and security standards the sector sets.
Financial services.
Later
Repeatable procedures around sensitive records and authorized transactions, with full audit and oversight.
Regulated industries need software that does the same thing every time and keeps a record of it. Deterministic, auditable execution is built for exactly that.Each deployment earns its sector's trust on its own terms.
Build the map. License the execution.
Workflow discovery is an upfront investment; validated execution is a recurring software capability. The commercial model covers institution-specific implementation, ongoing licensing, workflow maintenance and expansion. Every new map lowers the marginal cost of the tasks it covers, and every new institution makes the infrastructure more useful to the organizations that depend on it.
Gravitas is building the layer between AI reasoning and operational execution.