Innovation in the work.
The model is one part of AI. We put its power into the applications and workflows where people make decisions.
Exiled.AI / About us
AI belongs in the work.
Exiled.AI builds AI-native applications for financial markets. We put intelligence inside the applications where people make decisions—where it can see the work, show how it reached a result, and leave people in control. Exiled Markets is our first product.
“AI should build with you, show its work, and leave you in control.”
Our view of AI
Bullish on AI.
Vigilant with its power.
AI can sharpen decisions in finance, accelerate medical progress, extend our reach into space, and help answer society’s most vexing questions. We are building for that future.
Power at that scale deserves more than good intentions. As agents gain the ability to act, the systems around them must make authority clear, evidence inspectable, and human correction possible. That is what a responsible harness is for—and why we are building one.
What guides us
The work we believe in.
Exiled Markets starts with financial markets. These principles shape everything we build.
Talent in every form.
Different experiences, disciplines, and ways of thinking make stronger teams and better products.
Trust is engineered.
Show inputs, methods, and limits so people can inspect, challenge, and improve the result.
Progress for people.
We build for a future where more people can thrive—and take seriously the responsibility that comes with more capable AI.
Exiled.AI / The company
Cohesion means less software in every application.
Exiled.AI is an applied AI company building a model-independent platform for developing and operating AI-native applications.
Our platform brings computation, real-time data, interactive interfaces and governed agents into one coherent runtime. Applications share a common foundation for memory, logging, error handling, events and UI, so each one carries less of its own software, and less work to keep it running.
Smaller applications. Fewer integration failures.
Conventional stacks require developers to translate data, synchronize state and reconcile behavior across frameworks. Exiled’s components share schemas and runtime conventions. More functionality lives in the platform, leaving each application with less code to build, test and maintain.
A connected experience for people and agents.
The analytical engine, interface and agents work from the same application state. Real-time flow connects incoming events to calculations and views. This reduces synchronization work and framework overhead while supporting responsive, interactive applications.
Useful answers without a new application project.
Agents can assemble declared analytical capabilities and interface components around a question. Even a one-off investigation benefits from infrastructure already built. Useful calculations and surfaces can then be adapted for other questions or developed into recurring applications.
Fewer dependencies to manage and secure.
The zero-dependency core reduces exposure to third-party package vulnerabilities, transitive dependencies and version conflicts. Common controls give developers fewer separately configured integration boundaries to secure and operate.
Lower costs throughout the application’s life.
Smaller application code, shared infrastructure and consistent operating conventions reduce development, integration, testing, deployment and maintenance work. Teams can deliver more functionality, support more applications and spend less time diagnosing failures between frameworks.
A representative application
Bond Book: cash flows, duration, convexity and key-rate risk on a live Treasury curve, with an agent that can work the page.
Bond Book’s own code
Median application
Shared platform
- Sources: the Treasury curve
- Finance: bond math
- Frame: tables of cash flows
- Charts, tables, metrics, controls
- Agent and tools
- Logging and errors
- Streaming to the browser
- 0 third-party packages
Lines of code excluding comments and docstrings, counted in the repository in September 2026: Bond Book (cart/bonds.py), the median of the 84 application files, and the 37 platform modules every application shares. The platform is written and tested once; each application pays only for its own lines.
Stated precisely
- Size
- The claim is about each application, not the platform: the platform is large so that applications can be small, and its cost is shared across every application built on it.
- Latency
- Exiled reduces framework and orchestration overhead. Overall response time still depends on the models, data providers and calculations a request uses.
- Security
- A zero-dependency core reduces dependency and integration exposure. It does not remove vulnerabilities in our own code or in the infrastructure around it, which are reviewed and operated as any software is.
Exiled.AI / Applied research
Give agents a better foundation to build on.
Our research connects agentic code generation to a zero-dependency Python runtime. Time-series, tabular, and factor analysis share one set of contracts with the interface, streaming, and tools. An agent has fewer interfaces to reconcile as it builds an application; the harness governs what it can do, and the result remains open to inspection.
01 / RUNTIME + GENERATION
One system to build with.
Native time-series, tabular, and factor methods sit alongside charts and controls. Agents compose them through one runtime without a scikit-learn dependency.
02 / HARNESS + MEMORY
Agents that remember and answer for their work.
The model-independent harness sets tool permissions and limits. Adaptive memory recalls useful context, accepts corrections, and keeps a history of changes.
03 / MATH + EVIDENCE
Numbers people can inspect.
Defined methods perform financial calculations; the agent shapes the analysis. Sources, assumptions, and limits remain available for review.
Founder
Carl Carrie
Founder & CEO, Exiled.AI
Carl’s career spans institutional trading and technology at JPMorgan, Bloomberg, and LSEG. He has led product-driven growth across trading and software businesses. He writes regularly about AI and Wall Street and holds a software patent in high-performance computing. He is also a proud father and a young grandfather.
That experience shapes Exiled.AI: agents should be able to work inside the application, with live context and clear authority, while generated application code uses defined analytics and keeps the evidence available for review.
Selected writing
The ideas in more detail.
Two recent essays from Carl on the changing shape of institutional trading and what it takes to make software genuinely AI-native.
A Tale that Explains Why the Market Has Moved Beyond the Order
From execution algorithms to a governed decision loop around signals, portfolio risk, and the trades that follow.
Read the essay ↗AI-Native — a Three-Part Test!
A practical test of whether agents are built into the system: structural dependence, machine legibility, and evidence produced during execution.
Read the essay ↗