> For the complete documentation index, see [llms.txt](https://documentation.virtuedeep.tech/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://documentation.virtuedeep.tech/whats-new/migrating-to-the-new-model.md).

# Migrating to the new model

What changed in the platform model, what it breaks, and what to change in your integration.

Several platform concepts changed together. Three of them break existing integrations. This page covers what moved, what it breaks and what to do about it.

Nothing has been switched off. Superseded endpoints are hidden from the published API reference but remain callable, so you can migrate at your own pace.

## Summary

| # | Change                                                   | Breaks your integration |
| - | -------------------------------------------------------- | ----------------------- |
| 1 | Feeds moved from project scope to tenant scope           | Yes                     |
| 2 | Three attribute types became one                         | Yes                     |
| 3 | Connecting an LLM is now a prerequisite                  | Yes, for new projects   |
| 4 | Logics can span many projects                            | No                      |
| 5 | Solutions became a first-class concept                   | No                      |
| 6 | Interaction channels gained inputs alongside outputs     | No                      |
| 7 | Logic conditions and actions were renamed in the product | No                      |

***

## 1. Feeds are tenant-level

Feeds used to belong to a single project. They now belong to your account, and `projectId` has been removed from feed request parameters and responses.

**What to change.** Drop `projectId` from feed calls. If you were creating the same feed once per project, create it once.

**What stays.** Project has not disappeared from feeding entirely. You can still filter feed and attribute listings by project, project domain or project function. The difference is ownership, not visibility: a feed belongs to the account and is *filterable* by project rather than *owned* by one.

## 2. One attribute, not three

Feeding content attributes, interaction attributes and triggered action attributes are gone. There is one **attribute**, and the type now lives on the actions inside it.

| Before                                              | Now       |
| --------------------------------------------------- | --------- |
| Content attribute types, content attributes         | Attribute |
| Interaction attribute types, interaction attributes | Attribute |
| Triggered action attributes                         | Attribute |

A single attribute can hold actions of several types at once — media, interaction and triggered action. A logic output names a type and an attribute, and only the actions of that type inside the attribute are eligible for delivery.

**What to change.** Replace calls to the five old collections with the [attributes collection](https://documentation.virtuedeep.tech/guides/using-the-public-apis/working-with-actions-and-attributes/working-with-attributes.md). Where you previously chose between three attribute kinds, you now create one attribute and let what you feed into it determine which outputs can use it.

**Watch for.** An attribute is created empty, carrying nothing but a name. Actions arrive afterwards through feeding. Code that expected creation to establish a type needs to stop doing that.

## 3. An LLM must be connected before feeding

The platform no longer ships with a model. Connect one to your account — either a custom endpoint or an AWS Bedrock model — before you feed anything.

Projects that hold a conversation with a real person also need a model assigned to them. Projects that run on virtual humans do not.

| Project purpose   | LLM required |
| ----------------- | ------------ |
| `General`         | Yes          |
| `Data_Collection` | Yes          |
| `Simulation`      | No           |
| `Simulation_bulk` | No           |

**What to change.** Add two steps ahead of your existing provisioning flow: connect a model, then assign it to the project. See [configuring LLMs](https://documentation.virtuedeep.tech/guides/using-the-public-apis/configuring-llms.md).

**Watch for.** A project accepts one model at a time, and assigning a model to a simulation project is rejected rather than ignored.

## 4. Logics can span many projects

A logic used to belong to exactly one project. A logic can now apply to several.

**What to change.** Nothing, if you are creating one logic per project — that still works. If you were duplicating the same logic across projects, you can now create it once and attach the projects to it.

**Watch for.** A logic is scoped either by projects or by a solution, never both. Clearing the project list removes the scoping rather than leaving the logic attached to nothing.

## 5. Solutions

Domain and function used to be two loose fields you filled in yourself. A **solution** now pairs one domain with one function and the model that runs them, as a catalogue entry the platform provides.

This gives project creation two paths:

| Path             | You supply                                                          |
| ---------------- | ------------------------------------------------------------------- |
| Open solution    | A solution identifier, which carries the domain, function and model |
| Private solution | The domain and function directly                                    |

**What to change.** Nothing. The private path is what you are already doing. The open path is new and optional. See [working with solutions](https://documentation.virtuedeep.tech/guides/using-the-public-apis/working-with-solutions.md).

## 6. Interaction channels have inputs and outputs

Channels used to describe how responses leave the system. They now describe both directions: **inputs** capture what comes into the platform, **outputs** deliver what it produces.

Outputs carry an identity, so a consumer receiving several responses can tell which output produced each one.

**What to change.** Nothing immediately. Existing channel configuration continues to work. Adopt output identities when you need to disambiguate responses at the receiving end.

## 7. Conditions became variables, actions became outputs — in the product only

The blueprinting vocabulary changed. What used to be called a logic *condition* is now a **variable**, of one of three types: user motion, environmental and user geotargeting. What used to be a logic *action* is now an **output**.

**What to change.** Nothing. This is a naming change in the product and the documentation. The API request and response field names still say `condition` and `action`, and they have not moved. Where the documentation says variable, the payload field is still a condition field.

The field names will be brought in line during a later API revision, which will be announced separately.

***

## Endpoints hidden from the API reference

The following are no longer listed in the published reference. They still work.

| Area                                                   | Superseded by                   |
| ------------------------------------------------------ | ------------------------------- |
| Content attribute types and content attributes         | The attributes collection       |
| Interaction attribute types and interaction attributes | The attributes collection       |
| Triggered action attributes                            | The attributes collection       |
| Project-scoped feed endpoints                          | Their tenant-scoped equivalents |

If you depend on one of these, plan to move. They are hidden because they describe a model that no longer holds, and hidden endpoints are not maintained.

## Getting help

If something in your integration does not map cleanly onto the new model, raise it with your Virtue contact rather than working around it. Several of these changes have a straightforward equivalent that is not obvious from the endpoint list alone.
