Agentic Interpreted Applications

Amy doesn’t write your application. Amy runs it.

Code generators turn your words into a program you then own, deploy, and fix. Amy turns them into a model you can read and approve, and an AI runs that model live, turn by turn, with an engine guaranteeing every change. There is no code behind your application: nothing to build, nothing to deploy, no screens to design.

Open Amy See a template

The Amy App composing a turn: what needs attention, three figures drawn from the rows, the two places in focus with their row actions, and the moves from here.

One application, three parts

What is an Agentic Interpreted Application?

You approve the model. An AI interprets it when an operation needs judgment. The engine decides whether the proposed change is allowed.

The model

The readable description of what people do, what is stored, what must hold, and who decides what.

The interpretation

An AI reads the approved model and the current rows when an operation needs interpretation.

The guarantee

The engine validates every proposed change against the model and commits it, or refuses.

The AI never writes. The engine does.

Instantly live

Approve the model and the application exists.

There is no build step and no deployment. The App runs from the approved model, at its own address, the moment the model is approved. Change the model, approve the change, and the running application has changed.

Available at once

Every intent has an App. Open it from Amy and it runs, for you and for the people you share it with. No server to set up, no release to ship.

The interface is generated

Nothing is designed and no frontend is built. Each turn, Amy generates the interface from the model and the rows in front of it: a table, a record, a board, a form, a decision.

Changes are live

An approved change is live on the next turn, interface included. The data stays: rows outlive the shape of the model.

No build. No deploy. No screens to design.

How it works

Amy builds the model, then the model becomes the application.

You describe, read, approve, configure, and run. Every step stays visible.

Describe

Chat with Amy in plain language. The intent tree grows as you explain what people do and what must hold.

Amy in the IDE: the conversation builds the intent tree on the left, one persona, flow, contract, capability and record at a time.

Read and approve

Read the tree, operations in causal order, and screens. Every change is a proposal you accept or reject.

The Linear+ view of a model: who starts it, how it runs, what gets done, what it uses and what comes out, as columns joined by typed edges.

Configure

Choose who fulfils each capability, which model judges, and where rows live for each intent and target.

The Config tab beside Linear+: each capability the model consults, who fulfils it for this target, and the one still unanswered.

Run

Open the App. Amy composes each guided turn from the approved model and the current rows.

See every step

A record opened as a page in the Amy App: its real fields, the note the model wrote, and the operations that act on it.

The App

The interface is generated, every turn.

Nothing is designed in advance. Amy generates what you see from the model and the current rows, and the view dissolves into the next as you move. How it runs, and what it means that an AI runs it

Composed turns

Amy streams each guided turn from the approved model and the data in front of it.

Detail pages use the record's real fields.
References resolve to labels and appear as pickers.

A composed turn opening one record as a page: its fields by shape, provenance on the values the model chose, and the operations that act on it.

Every record has a working surface

Search across resolved labels. Filter, sort, and page. Row actions come from the model's operations on that record.

The standard record surface: search across resolved labels, the record’s columns, and one action button per operation the model defines on that row.

The data chooses the view

Boards, timelines, figures, and a judgment element appear when the model and data call for them.

Figures composed from the rows: a count, a filtered count and an average, chosen because the question asked for numbers.

You can see who acted and what the model used

Act as a person the model names. Follow the activity feed. Open the model log to see each prompt, result, time, tokens, and estimated cost.

The model log: tokens in and out, the estimated cost, and every call listed with its model, duration, retries and what it applied.

Templates

Start from an application that already runs.

Open a ready-made AIM model in Amy, read it, and adapt it to the application you need.

Template 01

Name from owner

Description, records, and operations will appear here.

Template 02

Name from owner

Description, records, and operations will appear here.

Template 03

Name from owner

Description, records, and operations will appear here.

See templates

The developer option

The same approved model can become conventional software.

The same approved model can go to a developer agent. It connects over MCP or pairs as an orchestrator, builds conventional software from the model, and every change record travels with it.

For developers

The standard underneath

Amy writes and runs AIM.

The Agentic Intent Model is an open, versioned standard for describing what an application is meant to do. Current version: 5.7.

Read about AIM Fork it on GitHub

Pricing

Start with one model. Pay when Amy becomes part of how you work.

Explorer is free. Paid plans add more Amy messages, projects, developer connections, and shared workspaces.

See pricing

Open the App

Amy doesn’t write your application. Amy runs it.

Open Amy