A model for your model
Proxa is a persistent workspace for your team and Claude that works with the files and tools you already use.
A DIFFERENT WAY TO THINK ABOUT MODELS
Why not give the LLM a Model?
We have spent the last few years extending language models with the things they need to do useful work. We gave them tools so they could act. Skills so they could perform specialized tasks. Memory and retrieval so they could access information. Harnesses so they could operate reliably.
Our view is that there may be another important layer: not another AI model, but a Domain Model — an explicit, persistent, computational representation of the thing the language model is working on.
The LLM reasons. The Domain Model represents. The Harness connects them.
A language model is built to reason, interpret, synthesize, and generate. A Domain Model serves a different purpose. It represents what is already known about a subject in a structured and deterministic form. Once something is known, there is little benefit in asking the LLM to infer it again. Relationships can be explicit. Calculations can be deterministic. State can persist. Dependencies can be maintained. The language model can focus on what actually requires reasoning.
That is the combination we are exploring with Proxa.
Why we built it
We started building this because we kept running into the same wall. A fund’s portfolio, a firm’s client roster, a company’s supply chain — every domain worth modeling was too rich to fit in a prompt and too important to leave unstructured. Prompting harder wasn’t going to fix that. So we built the layer we thought was missing. Proxa is a workspace for Domain Models. It’s a harness that allows you to design, build, and operate a model.
A Harness for Modeling
Coding has been a useful example of what happens when AI is given a proper working environment. A coding model becomes much more capable when it can work against a persistent codebase, use tools, execute code, observe results, test its work, and iterate. Proxa applies a similar idea to domain modeling. A user begins by describing what they want to model. AI helps design the Domain Model: what matters, how the subject should be represented, what information is required, and what methods or computation are appropriate. It then helps build the Model from the sources and tools already in use. Excel can remain Excel. Files can remain in shared folders. Existing systems can remain systems of record.
The important information is encoded into a form that AI can work with directly, while the underlying tools and workflows do not have to change. Once built, the Model can be operated continuously.
Process, not prompts
A Domain Model should not only become useful when someone asks it a question. Proxa uses agentic, human-in-the-loop Loops to keep the Model current over time. As source information changes, the Model can update. Governance and validation can happen continuously. Exceptions can be surfaced for review. The reports, presentations, dashboards, and other artifacts that depend on the Model can stay aligned with its current state.
These are not simply fixed pipelines. A Loop can interpret change, determine what needs attention, propose an action, ask for judgment when required, and continue once that judgment is supplied. Security, permissions, review, sandboxing, and history define the boundaries within which AI can operate. The Model becomes a maintained computational system rather than a static artifact.
Once the Model exists
The interaction with AI changes. Someone can ask Claude, ChatGPT, Copilot, or another AI tool a question against the Model. They can run a scenario, recalculate a forecast, update a presentation, generate a dashboard, build a form, create a workflow, or build a new tool on top of it.
The interface can change. The output can change. The AI model can change. The Domain Model remains.
That means the work no longer needs to begin by assembling enough context for each task. AI can work against a persistent representation of the domain and use the relevant structure and computation as needed.
A general-purpose modeling environment
The strength of Excel is that it’s general-purpose — it can model a budget, a schedule, a portfolio, anything with rows and columns. The strength of an LLM chatbot is general-purpose too — it can reason about almost anything you bring to it. We think domain modeling should work the same way. Historically, building this kind of system for a complex domain required a bespoke combination of software, data engineering, modeling, workflow infrastructure, and AI expertise. We think AI now makes a general-purpose approach possible.
Someone should be able to begin with a simple statement: “This is what I want to model.” AI can then help design it, build it from what already exists, keep it current, and operate it alongside them. The user does not need to design that architecture first. They start with the domain.