Acore
One typed application model
Acore UI · native API · PostgreSQLThe same Kanban experience, built three ways. Explore what changes when UI, API, data and policy become one connected model for humans and agents.
// One language. Connected layers.
model Task {
title: String
dependencies: List<String>
}
page Home {
query tasks = Board.list_tasks(...)
view { TaskBoard(tasks: tasks.data) }
}Create a task. Add a prerequisite. Move it through the board. Search, filter and edit. Each demo runs its own implementation and persists to its own database.
One typed application model
Acore UI · native API · PostgreSQLA shared Dart codebase
Flutter UI · Serverpod · PostgreSQLA full-stack web application
React · Next.js API · PostgreSQLPublic sandbox boards are shared between visitors. Use sample data. The embedded previews are interactive; open a full website for the complete viewport. Native targets are compiler-checked in this example; browser responsiveness is not a device test.
AxiomCore brings application structure, runtime behavior and reviewable evidence into the same workflow. Here is where that can make a difference as teams and codebases grow.
UI, API, data, validation and policy share a typed vocabulary. An agent can follow a feature across its declared boundaries.
Compiler-derived IDs, typed relationships and bounded projections make focused context retrieval possible without loading the entire codebase.
Trace affected endpoints, projections, UI consumers and targets through recorded dependency paths. Coverage and unknowns stay explicit.
Studio tools discover legal edits, update supported nodes and compile candidates. Broader backend and schema edit families are on the roadmap.
Typed contracts and generated clients carry data shapes across boundaries, reducing the integration code an author has to maintain.
This Orbit implementation has 13 counted authored files and 898 lines, including tests. Flutter + Serverpod has 18 files and 2,281 lines.
Inspector can explain ownership, reads, writes and permissions using graph facts and evidence references instead of a guessed architecture.
Developers and agents inspect the same declarations and diagnostics. Intent stays reviewable in source rather than disappearing into a prompt.
The semantic tool protocol is designed for existing capable models. A specialized edit model can be evaluated later against the same checks.
Source and graph revisions let tools identify stale context and bind a proposed change to the application the agent actually inspected.
Structured diagnostics, schema catalogs and JSON inspection outputs let an agent work with compiler facts through repeatable tools.
Pinned package contracts and resolved imports let teams share supported types and interfaces without copying their implementation into every prompt.
Declare cache strategy, lifetime and identity. In Orbit, successful mutations invalidate the shared board cache.
The contract carries input/output types, bounded validation and named failures, giving clients a consistent success and error surface.
Queries and mutations connect UI state to the runtime so loading, failure and refresh behavior follow the declared operation.
Guards, declared effects and resource grants make access requirements visible to the compiler and execution host.
Supported extension profiles run behind a sandbox and capability broker. Authority comes from explicit grants rather than ambient access.
Network, storage and other host operations pass through declared boundaries. An extension cannot gain a capability just by requesting it.
Timeouts, retries and circuit behavior are declared where supported. Unsafe operations retain explicit replay and idempotency boundaries.
Declared policies, compiled relationships and observed execution are distinct evidence layers, making audits easier to follow.
The runtime handles supported cross-cutting behavior; domain algorithms and database invariants remain authored, typed or implemented in supported extensions.
Supported query lifecycles track request identity and cancellation so superseded work does not silently become current UI state.
Supported stream profiles describe event types and delivery policy, keeping streaming behavior visible alongside the operation that produces it.
Supported UI state-persistence profiles can retain selected application state. Durable offline writes remain a separate capability.
Acore targets the browser and native Android/iOS hosts from shared UI source. Orbit has compiler checks for all three targets.
The web host renders DOM content and supports ordinary page metadata. Indexable HTML gives content-oriented sites a path to web SEO.
A minimal full-stack app can organize its UI, API and schema into three primary source files, with manifests and host setup around them.
Describe supported PostgreSQL and SQLite schemas, constraints and relationships with a typed database authoring profile.
Schema comparison, migration planning and supported apply/history workflows make database changes reviewable alongside application changes.
Artifact hashes, signatures and pinned dependency locks connect consumers to an exact reviewed contract version.
Compile and package the application for its selected host; keep environment configuration and secrets separate from authored source.
Extract or consume supported backend contracts and use existing clients incrementally. A full application rewrite is not required.
Contract checks, scenarios, API tests and target checks can verify more than the UI. Orbit passed the same 44 graph checks in each stack.
Describe responsive rules, input semantics, labels and interaction state in the UI source, then check the selected target.
Supported components carry typed inputs, events and composition boundaries, helping teams build a coherent design system.
Run semantic comparisons and relevant checks locally, then attach exact revisions, impact and test evidence to a repeatable review workflow.
From live updates to helpful form feedback, the shared runtime handles common behaviors that every app needs. Teams declare the rules and keep their attention on the product.
These are platform capabilities. The Kanban demo uses a subset; availability varies by host and selected profile.
Stream new information into a screen as it arrives, on supported hosts.
Check declared field rules and explain what needs fixing before data is accepted.
Give the app a consistent way to show what went wrong and what can happen next.
Reuse saved results when the app’s declared policy allows it.
Refresh the relevant saved results after a successful change.
Let screens show when an action is working, has finished or needs attention.
Retry eligible requests under the app’s declared rules.
Stop requests that exceed their allowed time, so a screen can offer a next step.
Coordinate supported request lifecycles so superseded results do not take over.
Check the declared sign-in and permission requirements at the operation boundary.
Limit supported sandboxed extensions to the specific capabilities they are granted.
Retain selected state where the chosen UI host supports persistence.
Check incoming and outgoing data against the application’s shared contract.
Apply supported capacity and queue limits instead of accepting unbounded work.
Connect operations and failures with execution records that teams can inspect.
Small apps can keep their schema, API and UI in three primary source files. Typed manifests, tests and deployment configuration sit alongside them as the app grows.
table tasks {
title: Sqlite.Text()
}Describe the data and supported constraints.
model Task {
title: String(maxLength: 120)
}
endpoint create_task(...)Declare typed operations, failures and policies.
page Home {
query tasks = Board.list_tasks(...)
view { TaskBoard(...) }
}Connect the same contract to web and native UI.
Illustrative excerpts, not a runnable three-file bundle. Orbit includes additional models, persistence code, migrations and tests. Generated files and setup are excluded from the authored-source counts.
Every agent needs to find the right part, understand it, make a change and check the result. Acore adds a direct path to supported application nodes.
Illustrative workflows, not a speed or token measurement. Acore Studio semantic UI editing is a preview capability.
“Rename the New task button to Create task.”
Retrieve a node. Edit a property.
Find the button that opens the task editor.
Find the widget. Edit the source.
Find the widget that opens the task editor.
Find the component. Edit the source.
Find the component that opens the task editor.
A node carries its identity, properties and relationships. The agent can request the relevant context and propose a supported edit while the compiler checks the candidate. The change stays visible in source for people to review.
Simplified tool calls and source excerpts. Source-based agents can also use language servers and structured tools. This label change needs no database migration or backend change in any stack. Supported scalar UI edits are demonstrated here; broader schema and backend edit families have separate availability.
All three demos implement the same product. Compare how each stack organizes UI, data, runtime behavior and tooling. AxiomCore connects those declared relationships in one application model.
| Capability | Acore CONNECTED MODEL | Flutter + Serverpod | Next.js |
|---|---|---|---|
| Application source | Acore across declared layers | Dart UI + Dart server | TypeScript UI + server |
| Web + Android + iOS | Shared UI; native target hosts | Shared Flutter widgets | Web; native needs a separate approach |
| Web presentation | DOM-based web host | App-centric Flutter renderer | HTML, React and server rendering |
| Application impact | Compiler-derived cross-layer evidence | Project tooling and application conventions | Project tooling and application conventions |
| Agent editing | Typed node tools in Studio preview | Source and ecosystem tools | Source and ecosystem tools |
| Caching | Declared policies and invalidation | Serverpod cache + application logic | Framework and application caching |
| Extension authority | Sandbox + explicit capability grants | Application and platform choices | Application and platform choices |
| Release evidence | Pinned, signed contract artifacts | Build and deployment pipeline | Build and deployment pipeline |
Source counts use the same retained-authored-code boundary, including tests. Generated code, dependencies, assets and project setup are excluded. File count and line count describe footprint; they do not establish development speed or token savings. Broader semantic editing is a Studio preview/roadmap capability, rather than a claim that this benchmark used every tool.
Read the detailed benchmark and methodology ↗
Build with Acore today. Join the AxiomStudio waitlist for the agent workspace built around your application model.
