Case Study
TX-1 — Headless Enterprise OS
Agentic terminal for enterprise operations
Tauri v2 · React 19 · LangGraph · Google OR-Tools · FastAPI · SQLite
The problem
Enterprise analysts are expensive, skilled, and spend the majority of their time doing things a system should do for them.
Not because the tools don’t exist. Because the tools are built around navigation rather than execution. Find the screen. Open the module. Download the CSV. Identify the row. Fix the thing. Re-upload. Re-run. Forty-five minutes for a problem that should take thirty seconds.
The UI is not the problem. The paradigm is.
The insight
Every enterprise failure mode follows the same pattern: something breaks, a human is notified, the human investigates, the human proposes a fix, the human implements it. The system’s job, in almost every enterprise tool built in the last twenty years, ends at notification.
TX-1 is built on a different assumption. The system’s job is to detect the failure, reason through the cause, calculate the optimal fix, and bring that fix — validated, specific, and reversible — to the human for a single approval decision.
Detect. Analyse. Propose. Approve.
That’s the entire interaction model.
What it does
A supply chain analyst types:
The system detects the capacity constraint violation. OR-Tools calculates the optimal rerouting across available nodes. Claude reasons through the cost implications and flags any secondary constraints. A High-Density Action Card renders in the terminal showing the proposed change, the cost delta, the constraint proof, and two options: approve or dismiss.
The analyst hits Enter.
The pipeline updates. The action is logged with full reasoning history. The monthly savings report gains another entry.
Forty-five minutes becomes thirty seconds.
Architectural decisions
Why a terminal interface?
Not nostalgia. Density. A terminal renders more actionable information per square inch than any dashboard UI, removes the cognitive tax of spatial navigation, and creates a command language that expert users can internalise. The goal was “speed of a CLI, precision of a CAD tool.” A web dashboard cannot achieve that.
Why local execution?
Enterprise data doesn’t leave the firewall without a six-month security review. By running the intelligence layer locally — a Tauri desktop shell backed by a Python sidecar — TX-1 operates inside the enterprise perimeter without requiring cloud connectivity or IT sign-off. The clean room architecture means no production data is touched until a human approves the action.
Why OR-Tools alongside the LLM?
Language models reason probabilistically. Logistics optimisation requires mathematical certainty. The TX-1 architecture uses Claude for reasoning and natural language understanding, and Google OR-Tools for constraint solving and solution validation. The agent proposes. The solver proves. The human approves.
Why LangGraph for orchestration?
The pipeline isn’t linear. Real enterprise workflows cycle — a proposed fix might surface a second constraint, which requires re-solving, which might reveal a third. LangGraph’s stateful cyclic graph handles this naturally and produces an auditable history of every decision made, which matters when a CFO asks why the system moved $60k of purchase orders.
What I learned
Building TX-1 clarified something that years of UX work had obscured: the most important design decisions in enterprise software are architectural, not visual. The choice of where the intelligence boundary sits — what the system decides versus what it proposes versus what it executes — is a design decision with more impact on user experience than any colour palette or component library.
It also clarified the limits of solo execution at enterprise scale, which shaped how I think about where this work goes next.
Status and direction
TX-1 is in active development. The core agentic loop, data connectivity layer, artifact system, and scenario engine are complete. Phase 4 covers the marketplace protocol, billing infrastructure, and the public scenario library.
The long-term vision is a platform: TX-1 as the execution layer, SS-1 as the intelligence layer, connected by a manifest schema that gives both systems a shared understanding of your business context.