Applications from a question.
An agent writes application code from intent, assembling connected charts, tables, metrics, controls and evidence around the decision—and extends the application when the next question arrives.
Exiled.AI / Architecture
One architecture for AI-native applications.
Exiled turns intent and varied data—including real-time streams—into working applications with rich visualizations and analytical capabilities. A governed agent assembles the experience on a compact, self-contained runtime. Exiled Markets is our first expression of that architecture in finance.
See how it fits together ↘INTENT → APP
A response you can work with.
Intent“What happens to this portfolio if the front end moves?”
AppThe agent generates code that assembles metrics, charts, tables, and scenario controls into a live application.
HarnessAsk lives inside the app; governed tools and defined analytics support the next decision.
The design of Exiled.AI
Four strengths, one architecture.
One programming model connects agents, data, analytical methods, and browser views. Exiled Markets applies that foundation to investment work; the same architecture can support other data-intensive domains.
Ask lives in every application.
The agent works from the application’s live state, data and tools, so it answers about what is on screen and builds with the same context the person sees.
Clear authority, inspectable results.
A model-independent harness gives agents declared tools and limits. Defined analytical methods compute the numbers; sources, methods and assumptions stay open to review.
A runtime you can own.
The Python standard-library core uses no third-party math, dataframe, model-wrapper, server or UI framework, and each user runs a separate, self-contained instance with its own state.
A generation example
From a question to a working application.
The agent does not draw a picture of an answer. It composes declared capabilities and components into an application that keeps working after the reply.
1 / INTENT
“Compare QQQ with SPY this year, then show what a 50 bp rate rise does to both.”
2 / GOVERNED CALLS
tickerQQQ against SPY, year to datescenariorates +50 bp, both instruments
Arguments checked against what is registered before anything is read.
3 / COMPOSED
- Metrics: return, volatility, drawdown, active return
- Chart: cumulative return, same dates
- Table: scenario effect by instrument
- Control: the rate shock, linked to the table
4 / THE APPLICATION
Live on current marks, with the method and dates behind each figure. Move the shock and the table recomputes; ask again and the application grows.
Simplified for illustration.
Architecture / Three demonstrations
What the component names actually do.
Each demonstration has one purpose: follow a number to its source, see which calculation a change reaches, and watch an agent choose tools before you take over the view. All three use the transition-cost example.
Follow an input to its displayed result.
Every step keeps its unit and says where it came from, so the figure on screen can be traced back to the quote it started as.
- Close180.00 USDequities.prices · 29 Sep 2026, 16:00 ET
- × shares to sell1,200target weight × NAV ÷ price, whole shares
- = notional216,000.00 USDunit carried through
- × cost rate12 bp10 bp slippage (assumed) + 2 bp half-spread (observed)
- = cost, as displayed259.20 USDthe table cell links back to each line above
Change one input. See what recalculates.
Three kinds of input move on different clocks. Pick one:
A new price revalues the book, re-sizes the trades, and moves turnover and cost.An edited weight re-sizes the trades and moves turnover and cost; values stay as they were.A new slippage assumption moves the cost only; nothing upstream recalculates.
Ask, inspect the tools, then take over.
The agent picks from the tools the application exposes. Actions that change what a person holds are not in its reach.
“Compare the target with a 50/50 blend, same costs.”
- portfolio.savedread
- proposal.targetread
- transition_costcalculate
- save_proposalnot in reach
- stage_tradesnot in reach
Exiled.AI / Around the model
Applications inherit the infrastructure around the model.
Frontier models connect as endpoints. Exiled provides the surrounding system: an Agent Harness, adaptive memory, a common tool protocol, the application runtime and built-in controls for sensitive data. Agent behavior, application state and operations stay in one architecture as models evolve.
These capabilities integrate with Exiled’s analytical engines, real-time flows and native UI through shared schemas, logging, error handling and event processing, so agents, calculations and interfaces operate together without extensive custom integration.
The resultSmaller, more capable applications: faster development, fewer model-specific integrations, more consistent safeguards, and lower maintenance and operating costs. Teams choose among models without making each application responsible for the infrastructure around them.
Reasoning, tools, bounded workflows.
The harness coordinates reasoning, tool execution and bounded workflows, with limits, retained results and inspection.
Useful context, kept.
Memory preserves the context worth keeping, accepts corrections and keeps a history of what changed.
One way to reach capabilities.
A common tool protocol gives every model consistent access to the application’s capabilities.
Sensitive data, handled once.
Built-in controls standardize the handling of personal data, secrets and other sensitive information, reducing the protective code each application must implement.
Application taxonomy / source access
One language for many kinds of market work.
Applications are organized around the decision, not a vendor screen. The agent selects a defined capability; the source registry resolves how to obtain the data and carries context into the resulting view.
01Portfolio & risk
02Rates, credit & options
03Company research
04Macro, events & news
Describe a risk, security, transaction, or market event.
Choose the data shape and analytical method the application needs.
Route by capability, access mode, credentials, and source readiness.
Build linked views with available provenance, assumptions, and source timing.
CONNECTED DOMAINS
Market prices & reference · Filings & private markets · Economic series · News & video · Geographic, weather & seismic events
WAYS TO ACCESS
Query for a result · Subscribe to a supported stream · Discover and open a file or dataset
Live streaming applies to supported feeds; other sources are queried or polled. Access depends on provider availability, credentials, and data licenses. Missing or unadjusted data should be disclosed rather than silently substituted.
Inside an application
From request to governed application.
The agent generates application code against Exiled’s data, analytical, and UI components. Frame and GraphFrame provide connected analytical state; a generation token prevents an older preview from overwriting a newer request.
01 / INTENT + AUTHORITYBind the work
Identify the request, inputs, and permitted data and tools.
03 / ISOLATED EXECUTION
MetricChartDataTable
Linked views and controls run in the user’s own instance. Ask works with the active application.
ONE RUNTIME
Native UI, agents, analytics, and web delivery share one runtime, built on the Python standard library.
DEFINED ANALYTICS
Generated applications call defined analytical methods for established calculations and conventions.
CONTROLLED EXECUTION
Agent tool permissions and provenance support review; generation tokens keep older responses from replacing newer application state.
One runtime / Many applications
The domain changes. The foundation travels.
Agents are most useful when they can act on data, call defined methods, and build something people can inspect. Exiled brings those pieces together as reusable application infrastructure.
INPUT / 01Varied sources
Streams, datasets, files, APIs, and reference material.
INPUT / 02Human intent
A decision, question, task, or evolving workflow.
EXILED / AI-NATIVE RUNTIME
One governed assembly layer.
The harness scopes model actions. Agents compose approved data access, analytical capabilities, and interface components in a shared programming model.
TOOLS · POLICY · MEMORY · EVENTS · UIOUTPUT / 01Live analytical views
Linked charts, tables, controls, and calculations.
OUTPUT / 02Inspectable applications
Browser experiences with context, evidence, and a record of the run.
IN FINANCE TODAYExiled Markets
Market data, news, portfolio context, and defined financial analytics become interactive desk applications.
THE BROADER PATTERNBeyond a single vertical
Wherever teams must combine changing data, domain methods, and human judgment, the same runtime can host a different set of approved capabilities.
Have a data-rich workflow beyond Markets? We would welcome an architecture conversation.
Talk with us about the runtime ↗The whole system
One complete instance per user.
Each authenticated identity runs a separate Exiled server and application stack. A container packages the instance; the hosting provider runs it in a microVM. Market data and news reach that instance, while desktop, tablet, and phone browsers open the same application. Other users have separate instances and caches.
Market dataLive prices · reference · licensed feeds
Economic + alternativeEconomy · events · seismic
NewsCurrent headlines · economic · event
IDENTITY A / SELF-CONTAINED INSTANCE
ONE PYTHON SERVER / COMPLETE EXILED RUNTIME
APPLICATIONS / AI INSIDEMarkets on the glass
Ticker · Portfolio · Options · Risk
ASK IN EACH APPAgent harness
Active app context · governed tools · adaptive memory · model-independent
RUNTIMEUI · Frame · GraphFrame
Charts and components
GOVERNANCEAuthority · provenance
Defined contracts
DELIVERYWeb · streaming
Behind an authenticating edge
STATE AT REST / Tape + SQLite · owned by this identity
No shared state
IDENTITY B / SEPARATE INSTANCE
Its own complete instanceApplications with Ask, governed runtime, server, and state.
The application adapts to desktop, tablet, and phone. Pure web. Nothing to install.
DesktopFull analytical workspace
TabletConnected views, on the move
PhoneThe decision in hand
Exiled Markets / Built on Exiled.AI
Bring us a workflow that needs more.
If your team has a workflow that today’s tools cannot reach, bring it. We will show how the architecture handles it.
Bring us a workflow that needs more Launch Exiled Markets ↗