> ## Documentation Index
> Fetch the complete documentation index at: https://wiki.agnes-ai.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Agnes Sys1 1.0

> Generally available Agnes Sys1 1.0 decision model: structured decision requests, response fields, pricing, and internal benchmark estimates.

> [!NOTE]
>
> Agnes Sys1 1.0 is generally available through the Agnes AI API. Access and rate limits depend on your account and API key permissions. Responses below are format examples.

## Overview

Agnes Sys1 1.0 is an Agnes AI decision model. Describe the current situation in `state` and define the questions in `questions`. The response returns structured results under matching keys in `answers`, rather than a chat completion.

The example in this guide routes a support request to a team and separately evaluates whether the user requests urgent handling. Other possible applications include workflow routing, intent classification, and business-rule screening.

| Item | Description |
| - | - |
| Model | `agnes-sys1-1.0` |
| Release status | Generally available |
| Method | `POST` |
| Endpoint | `https://apihub.agnes-ai.com/v1/systemone` |
| Input | A state string and named question definitions |
| Output | Structured decision results in `answers` |
| Billing | Input tokens are charged; output is free |

## Internal benchmark estimates

> [!NOTE]
>
> These figures come from the team's internal evaluation summary, where all model results in the supplied column are marked as estimates. They are not independently verified measurements, official Artificial Analysis results, or a public leaderboard ranking. Benchmark versions, sample sizes, and detailed methods have not been published; real-world performance may differ.

| Benchmark | Internal estimate | Composite weight |
| - | -: | -: |
| HLE-MultiChoice | 39.0% | 25% |
| When2Call | 95.2% | 25% |
| WildGuardTest | 93.6% | 25% |
| Omniscience accuracy | 38.8% | 12.5% |
| Omniscience non-hallucination rate | 73.7% | 12.5% |
| Internal weighted composite | 71.0 / 100 | 100% |

The composite is a weighted calculation using the estimates above, rounded to one decimal place. It is not an official AA index score.

## Request headers

Use your Agnes API key. Do not place keys in client-side code or public repositories.

```text theme={null}
Authorization: Bearer YOUR_API_KEY
Content-Type: application/json
```

## Input parameters

The following fields are used in the examples on this page. Required-field validation, defaults, and size or quantity limits are subject to the current service specification.

| Field | JSON type | Description |
| - | - | - |
| `model` | string | Model identifier: `agnes-sys1-1.0`. |
| `state` | string | The situation or text to evaluate. |
| `questions` | object | Named question definitions. The example evaluates `route` and `urgent` together. |
| `questions.<name>.type` | string | Question type. This guide demonstrates `choice` and `noul`. |
| `questions.<name>.instructions` | string | The instruction for evaluating this question. |
| `questions.<name>.criteria` | object | In the `choice` example, maps option keys to descriptions. |

### Choice question

Use `type: "choice"` and provide candidate options in `criteria`. In this example, the option keys are `billing`, `technical`, and `sales`. The selected key is returned in `answers.route.choice`.

### Numeric question: `noul`

The example uses `type: "noul"` for an urgency question and receives a numeric value in `answers.urgent.noul`. Keep the type string exactly as `noul`.

> [!NOTE]
>
> Preserve the numeric `noul` result returned by the service. Do not infer a calibrated probability, an allowed range, or a fixed business threshold from a single example; validate your decision rules against the production specification.

## Create a decision request

Use the dedicated `POST /v1/systemone` decision endpoint with the following request body. It is not the Chat Completions endpoint.

```bash theme={null}
curl --request POST 'https://apihub.agnes-ai.com/v1/systemone' \
  --header 'Authorization: Bearer YOUR_API_KEY' \
  --header 'Content-Type: application/json' \
  --data '{
  "model": "agnes-sys1-1.0",
  "state": "User feedback: The invoice amount is incorrect after upgrading my plan. Please investigate as soon as possible.",
  "questions": {
    "route": {
      "type": "choice",
      "instructions": "Which team should handle this request?",
      "criteria": {
        "billing": "Billing and payment support",
        "technical": "Technical support",
        "sales": "Sales inquiries"
      }
    },
    "urgent": {
      "type": "noul",
      "instructions": "Does the user ask for this to be handled as soon as possible?"
    }
  }
}'
```

## Response example

The following illustrates the response format. Decision results are returned under `answers`; the response is not a Chat Completions `choices` object.

```json theme={null}
{
  "model": "agnes-sys1-1.0",
  "answers": {
    "route": {
      "type": "choice",
      "choice": "billing",
      "probabilities": {
        "billing": 0.9981447473373126,
        "technical": 0.000863275841253645,
        "sales": 0.0009919768214337178
      },
      "confidence": 0.9972171210059689
    },
    "urgent": {
      "type": "noul",
      "noul": 0.9988554807973075
    }
  },
  "usage": {
    "input_tokens": 164,
    "output_tokens": 0
  }
}
```

## Response fields

| Field | JSON type | Description |
| - | - | - |
| `model` | string | Model identifier returned by the service. |
| `answers` | object | Named results matching the request's question names. |
| `answers.route.type` | string | The example returns `choice`. |
| `answers.route.choice` | string | The selected option key: `billing` in this example. |
| `answers.route.probabilities` | object | The probability values returned for the candidate keys. |
| `answers.route.confidence` | number | The confidence score returned for the choice result. Preserve the value returned by the service. |
| `answers.urgent.type` | string | The example returns `noul`. |
| `answers.urgent.noul` | number | The numeric result returned for this question. |
| `usage.input_tokens` | integer | Reported input-token usage; `164` in this example. |
| `usage.output_tokens` | integer | Reported output-token usage; `0` in this example. |

Read each result using its question name: `answers.route.choice` gives the selected team, while `answers.urgent.noul` gives the numeric result. Do not parse a prose answer or a `message.content` field.

The example selects `billing`, which corresponds to “Billing and payment support.” The confidence score is a separate returned field; do not substitute the largest option probability for it.

## Pricing

All prices below are in USD. `1M tokens` means one million tokens.

| Billing item | List price |
| - | -: |
| Input tokens | `$0.03 / 1M tokens` |
| Output | Free |

Input charge = reported input tokens ÷ 1,000,000 × `$0.03`

Structured decision output is free. The zero output-token count in this sample is a reported usage value, not a reason to ignore the `answers` object.

> [!NOTE]
>
> The input-token counting scope, account availability, and final billing details are subject to the production service specification and account bill.

## Integration checklist

* Use `agnes-sys1-1.0` in `model`.
* Send `state` and `questions`; do not replace them with chat-model fields such as `messages`.
* Keep question names consistent with the result keys under `answers`.
* Preserve the type names `choice` and `noul`.
* Retain `usage` when recording request usage.
* Confirm your account permissions and validate your numeric decision rules against the current service specification.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.