> 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/protocol/v1/models-and-profiles.md).

# Models and Profiles

Registration, versioning, execution requirements, and lifecycle of AI model profiles.

TrueOpen separates a **Model**, which identifies a model family, from a **Profile**, which freezes one executable and verifiable configuration of that model.

Every Task selects exactly one `model_id` and `profile_version`. That binding remains unchanged through inference, verification, challenge rounds, settlement, and evidence cleanup.

## What a profile defines

A profile must provide enough information for independent operators to reproduce execution and verification boundaries. Its committed manifest includes or binds:

* Model artifacts and content digests
* Tokenizer identity
* Supported Task types and generation mode
* Runtime class and required capabilities
* Input and output schemas
* Verification algorithm, evidence schema, and thresholds
* Resource requirements and resource tier
* Minimum ServiceBond
* Pricing and Task-limit inputs
* Challenge-window and timeout inputs

Display metadata, logos, README content, and installation advice may exist without becoming consensus fields. They cannot override the on-chain profile projection.

## Registration and versioning

Profiles under a Model use monotonically increasing versions. Registering a new version creates a new immutable profile record; it does not overwrite the old version.

A new version is required whenever a change affects execution, verification, canonical encoding, artifacts, tokenizer behavior, schemas, resource classification, pricing boundaries, or minimum bond. Existing Tasks continue using the snapshot bound when they were accepted.

The profile manifest is content-addressed. Implementations must verify that retrieved manifest material matches its committed hash and the registered projection before using it. Unknown manifest fields or encoding versions fail closed rather than being silently ignored.

## Resource tier and minimum bond

The resource tier describes the Profile, not a node's self-reported hardware. It is derived from factors such as compute scale, memory, context length, precision, modality, runtime constraints, and verification cost.

The minimum ServiceBond is also a Profile-level fact. Candidate selection, Task liability, and related economic checks use the same snapshotted value. Separate components must not invent different minimum-bond values for the same Profile.

One Cortex Node may support multiple Models with one ServiceBond. Model-level support uses the Model's support minimum; each Task also checks the selected Profile's minimum bond, capability, and capacity requirements.

## Profile lifecycle

A newly registered Model and Profile start in a registered state. They can accept initial Tasks through the defined cold-start rules, but support declarations alone do not prove active service.

Successful qualifying duties activate support for the corresponding operator and Model. Active support is tracked once per `(operator_address, model_id)`, even when that operator serves several Profile versions or both Task duties. Sufficient active Model-level support moves the Model to `ACTIVE`; a Profile remains `REGISTERED` and never receives an `ACTIVE` status. The operator must still be able to run the exact version frozen into each Task.

Governance can freeze, unfreeze, or delist a Profile according to the governance rules:

* A freeze blocks new Tasks while preserving deterministic handling of Tasks already in flight.
* Unfreezing restores future admission without rewriting history.
* Delisting is terminal for new Task admission.

Changing a display field is not a reason to mutate protocol identity. Changing anything that affects Task execution or judgment requires a new Profile version.


---

# 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/protocol/v1/models-and-profiles.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.