The specification stays.
The stack is replaceable.
Every coding agent writes code today. The problem comes after: what the software should do then lives only in the code of one particular framework, and with every change, code and specification drift further apart. With Amy, your team records what the software should do, independent of language, framework and agent. Your agent builds it in your stack and reports back which place in the code realizes which part of the model. That keeps drift checkable – and switching stacks doesn’t cost you the knowledge.
- Independent of language and framework
- For any coding agent with MCP
- Drift checkable, place by place
# purchasing.aim ## Operation: Approve ### Summary Approves a larger order. ### Expects - The order is over €2,500. ### Bindings - binds: `src/purchasing/approve.ts#approve` - binds: `route:POST /api/orders/:id/approval` - binds: `test/purchasing/approve.test.ts` The code may change. The rule stays, and the bindings show where it is realized.
The model knows no stack.
AIM describes what the software does: roles, data, rules and flows. Your agent chooses language, framework and database while building, and records the choice. The knowledge stays in the model, not in the code.
Switch stacks.
The same model can be realized again in another stack. Requirements and test cases come from the model again.
Switch agents.
Any coding agent that speaks MCP or reads an AGENTS.md works with the same model.
Open format.
AIM is an open specification. The model lives as readable files in your repository and belongs to you.
Code and model don’t drift apart unnoticed.
Every part of the model points to the place in the code that realizes it. A review compares the two, place by place, and reports what no longer fits.
Is something missing?
A rule from the model doesn’t appear at its place in the code, or a part has no realization yet.
Does something differ?
The code connects different things than the model says.
Is something orphaned?
A reference points to code that no longer exists.
Does the code do more?
Behavior that appears nowhere in the model. It becomes a question for your team, not a silent decision.
How you work together.
The specification doesn’t sit in a ticket. It is the model your agent works from.
Agree.
Settle behavior, responsibilities and exceptions with the people who know the work. Amy makes gaps visible before any code exists.
Build.
Your agent builds in your stack and reports back what it realized where. Code, tests and deployment stay with you.
Change.
Accepted changes arrive as change records. Your agent updates exactly the affected code and checks it against the new model.
Connect your agent.
Create a token under Developer access in Amy. Share gives you the ready-made command; the token is shown once.
No MCP? Export the model as a ZIP from Share into your repository. Its AGENTS.md carries the instructions for your agent. You create the tokens and decide which agent can access what. For your own build environment, there’s a self-hosted build runner.
claude mcp add --transport http amy "https://amybuilds.com/mcp" --header "Authorization: Bearer <TOKEN>"
- get_project
- Read the project and current editable version.
- list_versions
- List the project’s sequential versions.
- get_intent
- Pull the full approved AIM tree.
- get_change_records
- Pull accepted changes, also from a given date.
- get_content
- Pull supplied screen copy and starting rows.
- get_bootstrap
- Read the developer workflow and specification location.
Already have software? Amy reads it back.
Import a codebase or an AIM tree. Amy’s Encoder derives a draft model that your team reviews and confirms. From then on, you work on the model.
About AIMTwo starting points to try.
Copy a prompt into a new software project with Amy. Amy walks you through the open questions.
Customer service portal
Help me model a customer service portal for our own web application. Customers should submit requests and see their status. Our service team should assign owners and record responses. Ask me about roles, permissions, notifications, exceptions and the systems we need to connect. Help us review the intent before our coding agent implements it.
Field service app
Help me model a mobile app for field service technicians. They need assigned visits, checklists, notes and photos, and a way to record completed work. Ask me about offline use, synchronization, customer information, access and supervisor review. Make unresolved decisions visible so we can agree the intent before our coding agent builds the app.

What should your software do?
Start with what it should do. Your agent writes the code after that.
Tell Amy what your software should do→