# InlineAiModelDefinition

A model you define in full when you start a run, for one that isn't in Maitento's catalogue, such as a model on your own endpoint. Nothing is stored: only that run uses it. Its key and endpoint are not here: give them in the start request's `credentials`, with a `provider` that matches this model's `credentialProvider`.

**Kind:** an object, used in requests and responses. In a request it is called `InlineAiModelDefinitionInput`.

## In a request

| Field | Type | Required | Description |
|---|---|---|---|
| `name` | `string` | yes | A display name for the model, shown in logs and token usage. Constraints: required (empty or whitespace-only strings are rejected). |
| `apiType` | `string` | yes | The protocol the endpoint speaks: `"OpenAi"` or `"Anthropic"`. Constraints: required (empty or whitespace-only strings are rejected). |
| `modelIdentifier` | `string` | yes | The model id sent to the endpoint, exactly as the provider expects it. Constraints: required (empty or whitespace-only strings are rejected). |
| `credentialProvider` | `string` | yes | A name of your own for this model's key: the start request's `credentials` whose `provider` matches it are used. A model defined inline never falls back to Maitento's own key. Constraints: required (empty or whitespace-only strings are rejected). |
| `contextWindowTokens` | `integer` |  | The model's context window, in tokens. Constraints: range: 1..2147483647. |
| `maxOutputTokens` | `integer` |  | The most the model may write in one reply, in tokens. Constraints: range: 1..2147483647. |
| `supportsJsonSchemaResponseFormat` | `boolean` |  | Whether the endpoint accepts `response_format=json_schema`. False makes Maitento use `json_object` and restate the schema in the prompt instead. Default: true; send the real answer for your endpoint. |
| `tokenizerFamily` | `string \| null` |  | Which tokenizer to estimate the model's token counts with. Null: worked out from the model's name. |
| `reasoningEffort` | `string \| null` |  | The `reasoning_effort` sent with every request. Set it for a reasoning model whose own default doesn't work with tool calls, since every interaction sends tools. Null: not sent. |

```json
{
  "name": "...",
  "apiType": "...",
  "modelIdentifier": "...",
  "credentialProvider": "...",
  "contextWindowTokens": 0,
  "maxOutputTokens": 0,
  "supportsJsonSchemaResponseFormat": false,
  "tokenizerFamily": "...",
  "reasoningEffort": "..."
}
```

## In a response

Every field is present in a response; empty ones are `null`.

| Field | Type | Description |
|---|---|---|
| `name` | `string` | A display name for the model, shown in logs and token usage. Constraints: required (empty or whitespace-only strings are rejected). |
| `apiType` | `string` | The protocol the endpoint speaks: `"OpenAi"` or `"Anthropic"`. Constraints: required (empty or whitespace-only strings are rejected). |
| `modelIdentifier` | `string` | The model id sent to the endpoint, exactly as the provider expects it. Constraints: required (empty or whitespace-only strings are rejected). |
| `credentialProvider` | `string` | A name of your own for this model's key: the start request's `credentials` whose `provider` matches it are used. A model defined inline never falls back to Maitento's own key. Constraints: required (empty or whitespace-only strings are rejected). |
| `contextWindowTokens` | `integer` | The model's context window, in tokens. Constraints: range: 1..2147483647. |
| `maxOutputTokens` | `integer` | The most the model may write in one reply, in tokens. Constraints: range: 1..2147483647. |
| `supportsJsonSchemaResponseFormat` | `boolean` | Whether the endpoint accepts `response_format=json_schema`. False makes Maitento use `json_object` and restate the schema in the prompt instead. Default: true; send the real answer for your endpoint. |
| `tokenizerFamily` | `string \| null` | Which tokenizer to estimate the model's token counts with. Null: worked out from the model's name. |
| `reasoningEffort` | `string \| null` | The `reasoning_effort` sent with every request. Set it for a reasoning model whose own default doesn't work with tool calls, since every interaction sends tools. Null: not sent. |

```json
{
  "name": "...",
  "apiType": "...",
  "modelIdentifier": "...",
  "credentialProvider": "...",
  "contextWindowTokens": 0,
  "maxOutputTokens": 0,
  "supportsJsonSchemaResponseFormat": false,
  "tokenizerFamily": "...",
  "reasoningEffort": "..."
}
```

## Where it's used

- Inside [`AiModelOverride`](https://docs.maitento.com/reference/types/ai-model-override.md)
- Inside [`AiModelOverrideRequest`](https://docs.maitento.com/reference/types/ai-model-override-request.md)