# One canonical model.

> Rangoon is being designed around a structured intermediate representation that preserves agent capability meaning while adapters generate target-specific artifacts.

Canonical: https://rangoon.ai/architecture/

Status: In active development. Product screenshots are design previews with illustrative data. No general availability is implied.

## Model the capability

A canonical representation can include identity, activation, instructions, inputs, outputs, tools, dependencies, resources, permissions, policies, compatibility, provenance, and tests.

### Targets are outputs

Harness-specific files become compilation targets rather than the permanent underlying source of truth.

## Preserve provenance

Imported assets can retain repository, file, revision, lines, detection method, confidence, user modifications, ancestry, generated targets, and deployment history.

### No unexplained blobs

A capability should retain the evidence needed to understand where it came from.

## Keep authority external to configuration

The architecture separates composition and management from LNSAT evaluation for consequential execution.

### Design in progress

This architecture describes planned product direction rather than a stable public contract.


## More information

- [Documentation index](https://rangoon.ai/llms.txt)
- [AI access and policies](https://rangoon.ai/ai/)
- [Source repository](https://github.com/hypler-dev/rangoon)