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

# Getting Started

> Submit your first evaluation task in under 5 minutes

## Quick start

Submit an evaluation task via email or API and start testing your AI agents.

<Tabs>
  <Tab title="Email Method">
    ### Email submission (easiest)

    Send an email to test your first evaluation:

    ```
    To: agent@mixus.com
    Subject: Eval: Calculate Sales Commission

    Calculate 15% commission on a $50,000 sale.

    Test type: with-verification
    Auto-detect: true
    Reviewer: your-email@company.com
    ```

    **What happens next:**

    1. System creates an agent to solve the task
    2. AI automatically detects verification checkpoints
    3. Agent executes and pauses at checkpoint
    4. You receive email notification
    5. Reply "approve" to continue
    6. Get results email when complete

    <Check>
      **No API key needed** - Just send an email!
    </Check>
  </Tab>

  <Tab title="API Method">
    ### API submission (for automation)

    #### Step 1: Get API key

    Generate an API key at [app.mixus.ai/integrations/api-keys](https://app.mixus.ai/integrations/api-keys) with `eval:create` and `eval:read` permissions.

    #### Step 2: Submit task

    ```bash theme={null}
    curl -X POST https://app.mixus.ai/api/eval/create-task-agent \
      -H "Authorization: Bearer mxs_eval_YOUR_KEY" \
      -H "Content-Type: application/json" \
      -d '{
        "taskName": "Calculate Sales Commission",
        "taskDescription": "Calculate 15% commission on a $50,000 sale",
        "autoDetectCheckpoints": true,
        "testMode": "with-verification",
        "assignedReviewer": "your-email@company.com"
      }'
    ```

    #### Step 3: Track progress

    ```bash theme={null}
    # Use executionId from response
    curl https://app.mixus.ai/api/eval/status/EXECUTION_ID \
      -H "Authorization: Bearer mxs_eval_YOUR_KEY"
    ```
  </Tab>
</Tabs>

***

## Test modes explained

Choose the right mode for your evaluation:

<AccordionGroup>
  <Accordion title="With Verification + Auto-detect (Recommended)">
    AI automatically determines where verification is needed.

    **Use when:**

    * You don't know exact verification points
    * Task involves external actions (emails, purchases)
    * You want AI to identify critical steps

    **Example:**

    ```json theme={null}
    {
      "testMode": "with-verification",
      "autoDetectCheckpoints": true
    }
    ```
  </Accordion>

  <Accordion title="With Verification + Manual Checkpoints">
    You specify exact verification points.

    **Use when:**

    * You know exactly where verification should happen
    * Testing specific decision points
    * Task has well-defined stages

    **Example:**

    ```json theme={null}
    {
      "testMode": "with-verification",
      "checkpoints": [
        {
          "stage": "calculation",
          "description": "Calculate result",
          "verificationQuestion": "Is the calculation correct?"
        }
      ]
    }
    ```
  </Accordion>

  <Accordion title="Without Verification (Baseline)">
    Agent runs autonomously without human oversight.

    **Use when:**

    * Speed benchmarking
    * Comparing with/without human
    * Low-risk tasks

    **Example:**

    ```json theme={null}
    {
      "testMode": "without-verification"
    }
    ```
  </Accordion>
</AccordionGroup>

***

## Verification workflow

When using verification mode, here's what to expect:

<Steps>
  <Step title="Checkpoint reached">
    Agent pauses execution and sends email notification
  </Step>

  <Step title="Review agent's work">
    Open chat or check email to see what agent plans to do
  </Step>

  <Step title="Respond">
    * Type **"approve"** - Agent continues - Type **"reject"** - Agent stops -
      Type **"hint: \[guidance]"** - Agent adjusts approach
  </Step>

  <Step title="Agent continues">
    After approval, agent proceeds to next step or completes task
  </Step>
</Steps>

***

## Example: Complete flow

Here's a complete example from start to finish:

### 1. Submit via email

```
To: agent@mixus.com
Subject: Eval: Research Competitors

Research pricing for top 3 AI agent companies and email
summary to team@company.com

Test type: with-verification
Auto-detect: true
Reviewer: manager@company.com
```

### 2. Confirmation received

```
From: agent@mixus.com
Subject: Re: Eval: Research Competitors

✅ Evaluation Started!

Task: Research Competitors
Type: with-verification
Checkpoints: 2 (AI-detected)

Review at: https://app.mixus.ai/chat/abc123

You'll receive email when verification is needed.
```

### 3. Checkpoint notification

```
From: agent@mixus.com
Subject: Checkpoint Verification Needed

🔔 Checkpoint 1: Research Review

I've researched pricing for OpenAI, Anthropic, and Google DeepMind.

Results:
- OpenAI: $20-$200/month
- Anthropic: Custom enterprise pricing
- Google DeepMind: Part of Google Cloud

Respond with: approve | reject | hint: [guidance]

View full details: https://app.mixus.ai/chat/abc123
```

### 4. You approve

Reply to email: `approve`

### 5. Second checkpoint

```
From: agent@mixus.com
Subject: Checkpoint Verification Needed

🔔 Checkpoint 2: Email Review

Ready to send email summary to team@company.com

Subject: Competitor Pricing Analysis
Content: [Shows email draft]

Approve sending this email?
```

### 6. You approve again

Reply: `approve`

### 7. Completion notification

```
From: agent@mixus.com
Subject: Evaluation Complete

✅ Task Complete!

Results:
- Success: Yes
- Checkpoints: 2/2 approved
- Duration: 5 minutes
- Cost: $2.50

Email sent successfully to team@company.com

View full details: https://app.mixus.ai/chat/abc123
```

***

## Where to track evaluations

<CardGroup cols={2}>
  <Card title="Dashboard" icon="gauge" href="https://app.mixus.ai/eval">
    Visual overview of all evaluations
  </Card>

  <Card title="Chat Interface" icon="comments" href="https://app.mixus.ai">
    See full conversation and agent work
  </Card>

  <Card title="API Status" icon="code" href="/eval-api#get-execution-status">
    Programmatic status checks
  </Card>

  <Card title="Email Notifications" icon="envelope">
    Receive updates via email
  </Card>
</CardGroup>

***

## Tips for success

<Tip>
  **Write clear task descriptions**

  Good: "Calculate 15% commission on a \$50,000 sale and send result via email to [manager@company.com](mailto:manager@company.com)"

  Bad: "Do commission stuff"
</Tip>

<Tip>
  **Start with auto-detect mode**

  Let AI determine checkpoints until you understand the system better.
</Tip>

<Tip>
  **Test simple tasks first**

  Start with calculations or research before complex multi-step workflows.
</Tip>

<Tip>
  **Use baseline mode for comparisons**

  Run same task with and without verification to measure impact.
</Tip>

***

## Next steps

<CardGroup cols={2}>
  <Card title="Task Preparation" icon="pen" href="/evaluation/task-preparation">
    Learn how to write effective evaluation tasks
  </Card>

  <Card title="Examples" icon="lightbulb" href="/evaluation/examples">
    See 10+ ready-to-use example tasks
  </Card>

  <Card title="API Reference" icon="code" href="/eval-api">
    Complete API documentation
  </Card>

  <Card title="Best Practices" icon="star" href="/evaluation/best-practices">
    Tips for getting the most from evaluations
  </Card>
</CardGroup>

***

## Common questions

<AccordionGroup>
  <Accordion title="Do I need an API key for email submissions?">
    No! Email submissions don't require an API key. Just send to [agent@mixus.com](mailto:agent@mixus.com) with subject starting "Eval:"
  </Accordion>

  {' '}

  <Accordion title="How long does an evaluation take?">
    Depends on task complexity and verification time. Simple tasks: 2-5 minutes.
    Complex tasks: 10-30 minutes.
  </Accordion>

  {' '}

  <Accordion title="Can I submit multiple tasks at once?">
    Yes! Via API you can submit multiple tasks. They'll run in parallel.
  </Accordion>

  {' '}

  <Accordion title="What if I don't respond to a checkpoint?">
    The evaluation will wait for your response. You can respond anytime via email
    or chat.
  </Accordion>

  <Accordion title="Can I cancel a running evaluation?">
    Yes, reply "reject" at any checkpoint or stop it from the dashboard.
  </Accordion>
</AccordionGroup>

***

## Need help?

* **Email:** [support@mixus.ai](mailto:support@mixus.ai)
* **Dashboard:** [app.mixus.ai/eval](https://app.mixus.ai/eval)
* **API Keys:** [app.mixus.ai/integrations/api-keys](https://app.mixus.ai/integrations/api-keys)