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

# Copilots

Copilots are the chat applications end users open to use your recipes, with feedback, traces, file upload, voice, and conversation history.

Karini AI copilots are the end-user chat applications that use your recipes as the engine for conversations. You can design copilots that follow your organization's communication guidelines, represent your brand, and capture feedback from users.

For details on creating and configuring a copilot, refer to [Create Copilot](https://docs.karini.ai/recipes/workflow-recipe/create-copilot.md).

Key capabilities of Karini AI copilots:

#### **Seamless access**

Users can open copilots inside the Karini AI platform if they have the Copilot user, Administrator, or Superuser [role](https://docs.karini.ai/user-management.md#user-roles). You can also embed a copilot in your own domain so your domain users can access it there.

#### **Engaging in dialogue**

Users ask questions and receive responses in an interactive conversation. Copilots support streaming for almost all leading model providers, plus citations (references) and optional follow-up questions. This keeps the user experience consistent when you switch model providers for A/B testing or model upgrades.

#### **Feedback mechanisms**

Beyond basic upvotes and downvotes, copilots can capture several types of custom feedback:

* **5-stars:** users rate the response with a 5-star rating.
* **Options:** predefined options for users to choose from, configured for your requirements. For example: "Inaccurate," "Vaguely accurate," "Accurate."
* **Text feedback:** users write descriptive feedback.
* **Correction:** users suggest corrections or improvements to the response.

#### **Detecting greetings and non-specific questions**

Copilots can detect irrelevant or unrelated questions and respond with an appropriate message that keeps users engaged and encourages them to ask questions on the topic.

#### **Content safety filters**

You can enable [content safety filters](https://docs.karini.ai/recipes/knowledgebase-recipe/create-recipe.md#content-safety) in the recipe to prevent harmful or toxic content from being accepted or generated in the copilot, and respond to such content according to your organization's content safety messaging policy.

#### **Copy functionality**

Users can copy both questions and answers for reference and sharing.

#### Starting a New Conversation

The Start New Conversation option creates a new conversation thread, keeping multiple interactions separate for easier tracking and management. It includes:

* **Edit Thread Name:** rename a thread for easier identification.
* **Delete Thread:** delete a conversation thread when it's no longer needed.
* **Conversation List:** all conversation threads are listed in the left panel.
* **Search:** a search bar at the top helps users find a specific thread.

#### Smart Search

The Smart Search toggle turns smart search on or off.

<figure><img src="https://415930246-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F7ZrVuiAUMyuYVvrK5KaB%2Fuploads%2FEPjqsaeUWdnFTk2c5v29%2Fimage.png?alt=media&amp;token=dfc40a78-9f18-4197-b3b5-00d7344221cd" alt=""><figcaption></figcaption></figure>

* When enabled, the system uses a **Dynamic Filtering prompt** to construct queries, so search results are more relevant and contextually appropriate. When **Smart Search** is on, the **Metadata Filtering** component is also used, improving the efficiency of data retrieval and filtering.

<figure><img src="https://lh7-rt.googleusercontent.com/docsz/AD_4nXfpeJtHgOdeAlfYj97ou5dctLD0XJ7KIp-WPHp2eLRi0T81K5Gc_HzYsRqt7s8yi_qjHTwKlv2riT5_BiTuRnAsQrYbA5Z93nofMhqc9K-6AufPrk0yq7Lf-6-X5V8Y21aT5Iyb?key=hZ-bbCjTmw_ft_MIfRiZmVMj" alt=""><figcaption></figcaption></figure>

* When disabled, the system doesn't use dynamic filtering, and search falls back to a standard method. Results may be less contextually relevant, with no advanced filtering applied.

<figure><img src="https://lh7-rt.googleusercontent.com/docsz/AD_4nXc3fv1jsAVpn0sZ5UPy9W1DiU6tKExCFItAKGwNjlOuKl4ndn4FIiwLRn1MnxoE5CSuOAbl200GBWSf3xOUS7i4WD4NjTJIfd9QlyrMjjEkV0OQdHWrqNbvVQO9KCsEp2L6Oac8?key=hZ-bbCjTmw_ft_MIfRiZmVMj" alt=""><figcaption></figcaption></figure>

#### Attachments

The attachment icon lets users upload files to include in the conversation, such as documents, images, or other data that gives the copilot more context. The system processes attached files alongside the user's message for more informed responses.

{% hint style="info" %}
To upload files in a copilot, the Global Embedding Model must be selected on the [Organization](https://docs.karini.ai/organization.md) page.
{% endhint %}

#### **Tracing**

**Tracing Details** let users follow how each component ran in response to a query, showing which tools and processes were used. This gives transparency into how the agent works and how efficiently it operates.

<figure><img src="https://415930246-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F7ZrVuiAUMyuYVvrK5KaB%2Fuploads%2FjENHrvfDzmjVXZkRwVRd%2Fimage.png?alt=media&amp;token=bc35afc7-b245-4db1-8d06-1f88ab078535" alt=""><figcaption></figcaption></figure>

#### **Prompt Lens**

**Prompt Lens** shows the detailed prompt behind the agent's response, giving transparency into how the system processed the input and generated the answer, including data retrieval, analysis steps, and reasoning.

<figure><img src="https://415930246-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F7ZrVuiAUMyuYVvrK5KaB%2Fuploads%2FqpY0zt2jYkU9Xg1taGkw%2Fimage.png?alt=media&amp;token=f96ac091-4ecd-465a-aec2-219141bfceb2" alt=""><figcaption></figcaption></figure>

#### Audio: Voice Input and Output

Copilots support voice input and voice output. Users can speak their questions, which the system processes to generate responses, and the copilot can reply with audio. This enables hands-free interaction and helps in situations where typing isn't practical, or for users with accessibility needs.

<figure><img src="https://415930246-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F7ZrVuiAUMyuYVvrK5KaB%2Fuploads%2FdhNTd2Ac04lxGGeK7gew%2Fimage.png?alt=media&amp;token=44680722-9b1c-45b6-b399-1c782de64ba4" alt=""><figcaption></figcaption></figure>

#### Copilot History

You can access the full chat history of a copilot, which is useful for auditing, cost and usage monitoring, and performance improvement, and for spotting trends and usage patterns.

Together, follow-up questions, feedback, and corrections make copilots a foundation for continuous learning and improvement.

#### Copilot Actions

On the copilots list page, use the **Actions** button for the following actions. Only Administrator and Superuser [roles](https://docs.karini.ai/user-management.md#user-roles) can see and perform them.

#### **Edit Recipe**

Opens the copilot's underlying recipe, where you can adjust the workflow. You must publish the recipe for changes to appear in the copilot. You can also [create multiple copilots](https://docs.karini.ai/recipes/workflow-recipe/create-copilot.md) from the same recipe as your use case requires.

#### **Edit Export**

Opens the copilot's configuration, where you can edit its messages, layout, feedback settings, and other options. Save the configuration for the changes to apply to new sessions.

#### **View History**

Opens the Copilot History page, with a [**dashboard**](https://docs.karini.ai/dashboard-overview/copilot-dashboard.md) of copilot activity and a table of detailed information about each interaction:

* **Request Id**: a unique identifier for each request.
* **User name**: the person interacting with the copilot.
* **API Session**: the copilot session ID.
* **Recipe name**: the recipe behind the copilot.
* **Recipe version**: the recipe version used.
* **Chatbot name**: the copilot name.
* **Status**: the response status, such as success, aborted, unsafe, or greeting.
* **Question**: the user's question.
* **Audio Input**: whether audio input was provided.
* **Answer**: the copilot's response.
* **Error**: any error that occurred while processing the request.
* **Files**: files the user uploaded in the interaction.
* **Metadata**: additional context for the request, such as session details or user context.
* **Response Type**: the type of response, such as copilot.
* **Thread**: the conversation thread for the request.
* **Feedback**: feedback provided by the user.
* **Created**: when the request was created.
* **Updated**: when the request was last updated.
* **Tokens**: the number of tokens used to generate the response.
* **Updated By**: the person who updated the request.
* **Faithfulness**: how faithful the response is to the retrieved context.
* **Answer Relevancy**: how relevant the answer is to the question.
* **Audio Response**: whether the response was audio.
* **Origin Type**: the source of the request (user or system).
* **Bot Name**: the chatbot that handled the request.
* **User Name**: the user who made the request (in external chatbots).
* **Tenant Id**: the unique identifier for the tenant.
* **Violation Report**: any violations found during the interaction.

#### View Traces

For each record in the copilot history table, you can view traces showing step-by-step how the request was processed and the response generated. A trace has two sections, **Prompt** and **Attributes**.

1. **Prompt:** the trace of each operation executed during processing, including its input and output.
2. **Attributes:** parameters and metrics for the request, such as:
   * Input tokens
   * Completion tokens
   * Model parameters such as temperature and max tokens

<figure><img src="https://415930246-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F7ZrVuiAUMyuYVvrK5KaB%2Fuploads%2FfnX6D0Q4Uo4c4Xnnka9x%2Fimage.png?alt=media&amp;token=45d8ebdc-b44d-4b38-84dc-136445aa76da" alt=""><figcaption></figcaption></figure>

#### Export Dataset

Copilot history holds valuable information about user interactions, such as sessions, users, questions, answers, feedback, and prompts. You can select records and export them for downstream processing, auditing, or fine-tuning a model. For details, refer to [**Copilot History: Dataset Export**](https://docs.karini.ai/dataset-export.md#copilot-history-dataset-export).

#### Attachments & Integrations

In a copilot, the **"+" (Add)** button next to the message box opens the **context and integrations menu**, so users can attach content or connect external tools before sending a message. This enriches the conversation with additional data sources and improves response accuracy.

* **Open the Add menu:** click the **"+"** icon to see the available options.
* **Connectors:** select **Connectors** to see available integrations, including **Remote MCP Servers**.
  * If a remote MCP server isn't authorized yet, the copilot shows a **Connect** action and prompts the user to complete authorization.
  * Once connected, the server is available for the copilot to use in the conversation.
* **Add files or photos:** upload documents or images for the copilot to reference during the interaction.

<figure><img src="https://415930246-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F7ZrVuiAUMyuYVvrK5KaB%2Fuploads%2FMb3f5UQ1lOAbU7PSFQ5L%2Fimage.png?alt=media&amp;token=2646e735-36cc-4d08-90e7-ee7683cd2526" alt=""><figcaption></figcaption></figure>

#### Mass Feedback

In Copilot History, **administrators** can select **multiple conversation** records and submit feedback for all of them in one bulk action. If a selected record already has feedback, the bulk update overwrites it with the new values.

Bulk feedback supports:

* **Vote** (thumbs up or thumbs down)
* **Rating** (1 to 5)
* **Text feedback** (written comments)
* **Corrected response** (the expected answer, if applicable)
* **Additional custom fields** (any other configured attributes)

<figure><img src="https://415930246-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F7ZrVuiAUMyuYVvrK5KaB%2Fuploads%2FfjQCWrNKBcwUiTpbOFbw%2Fimage.png?alt=media&amp;token=65a77624-649c-440d-8470-db6ae98d5467" alt=""><figcaption></figcaption></figure>

#### Copilot File Upload: Supported Formats and Limits

You can attach files directly in a copilot conversation. Here are the supported formats, size limits, and upload behavior.

**General Upload Limits**

Files that exceed these limits are rejected before the upload starts.

| Constraint    | Limit              |
| ------------- | ------------------ |
| Max file size | 50 MB per file     |
| Max files     | 5 files per upload |

**Supported File Types**

* **Images**: common image formats, shown as inline previews in the chat.

| Format | Extension |
| ------ | --------- |
| JPEG   | .jpeg     |
| JPG    | .jpg      |
| PNG    | .png      |
| GIF    | .gif      |
| TIFF   | .tiff     |

* **Documents**: standard office and text documents, the most common formats for sharing reports, spreadsheets, and presentations.

| Format | Extension |
| ------ | --------- |
| PDF    | .pdf      |
| DOC    | .doc      |
| DOCX   | .docx     |
| XLS    | .xls      |
| XLSX   | .xlsx     |
| PPT    | .ppt      |
| PPTX   | .pptx     |
| TXT    | .txt      |
| CSV    | .csv      |

* **Audio**: recordings for transcription or analysis. Audio files are converted to Base64-encoded data before being sent to the server.

| Format | Extension |
| ------ | --------- |
| M4A    | .m4a      |
| MP3    | .mp3      |
| WAV    | .wav      |

* **Video**: common video formats, for when the copilot needs to process or reference video content.

| Format | Extension |
| ------ | --------- |
| MP4    | .mp4      |
| MOV    | .mov      |

#### Copilot Draft Mode: Document Upload

Copilot Draft Mode has its own upload option, separate from the chat attachment menu. Use it to load an existing document into the copilot's draft editor for review or editing. Only one file can be uploaded at a time.
