> For the complete documentation index, see [llms.txt](https://docs.trueopen.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.trueopen.ai/providers/model-integration/cortex-architecture.md).

# Cortex Architecture

Understand Cortex components, trust boundaries, and network interactions.

Cortex is the local control plane of a compute provider. It turns chain assignments and Builder coordination messages into constrained calls to a model adapter, validates the returned material, and signs protocol actions.

```mermaid
flowchart LR
    O[Compute provider] --> C[Cortex control plane]
    C -->|queries, events, self-rescue| K[TrueOpen chain]
    C -->|coordination and Task data| N[Builder Nexus]
    C -->|adapter operations| A[Model adapter]
    A --> R[Compute runtime]
    C --> S[(Task state and evidence)]
```

## Components and trust

| Component               | Trust role                  | Responsibility                                                                    |
| ----------------------- | --------------------------- | --------------------------------------------------------------------------------- |
| Cortex control plane    | Trusted                     | Keys, chain reconciliation, policy, Task state, signing, retries, and self-rescue |
| Model adapter           | Constrained                 | Converts a model runtime into stable inference and verification operations        |
| Compute runtime         | Untrusted input to Cortex   | Produces raw output, scores, and intermediate material                            |
| Task and evidence store | Recoverable local authority | Persists leases, signed material, deduplication records, evidence, and cursors    |

The compute runtime does not decide protocol verdicts, hold the service key, or communicate directly with the chain or Builders. Cortex checks returned bindings and digests before signing.

## Outbound connections

A Cortex Node initiates connections to the chain, Builder Nexus endpoints, its adapter, and operator-controlled management systems. It does not require a public inbound endpoint.

An event or Nexus message prompts reconciliation; it is not final authority. Cortex uses chain queries to determine whether assignment, submission, settlement, or finality has been accepted.

## Trust order

Cortex evaluates information in this order:

1. Chain query results
2. Chain events, which trigger a query
3. Nexus acknowledgements and coordination messages
4. Runtime observations and operator configuration

A lower level cannot override a higher one. For example, a successful Nexus response proves transport, not chain acceptance.

## What Cortex does not do

Cortex does not participate in consensus, choose network-wide Workers or Verifiers, decide the final Task verdict, calculate final payouts, or convert local queue and hardware claims into protocol facts.

For practical operation, use the [compute provider guide](/providers.md). To implement the runtime boundary, use the [model integrator guide](/providers/model-integration.md).


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter:

```
GET https://docs.trueopen.ai/providers/model-integration/cortex-architecture.md?ask=<question>&goal=<endgoal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.