> ## Documentation Index > Fetch the complete documentation index at: https://koreai.mintlify.app/llms.txt > Use this file to discover all available pages before exploring further. # Adding Custom Prompts [Back to Prompts Library](/ai-for-service/generative-ai-tools/prompts-library) Reference and procedures for managing Custom prompts in the Prompts Library. *** ## Add a Custom Prompt ### Prerequisites Integrate a pre-built or custom LLM before creating a prompt. See [LLM Integration](/ai-for-service/generative-ai-tools/llm-integration). ### Steps 1. Go to **Generative AI Tools** > **Prompts Library**. 2. Click **+ New Prompt** (top right). 3. Enter the **Prompt Name**, then select the **Feature** and **Model**. New prompt — name, feature, model 4. The **Configuration** section (endpoint URL, auth, headers) is auto-populated from the model integration and is read-only. Configuration section 5. In the **Request** section, create a prompt or import an existing one. Request section **To import an existing prompt:** 1. Click **Import from Prompts and Requests Library**. Import from Prompts and Requests Library 2. Select the **Feature**, **Model**, and **Prompt**. Hover and click **Preview Prompt** to review before importing. You can interchange prompts between features. 3. Click **Confirm** to import the prompt into the JSON body. **To create from scratch:** Click **Start from scratch** and enter the JSON request for the LLM. Start from scratch 6. (Optional) Toggle **Stream Response** to enable streaming. Responses are sent incrementally in real time instead of waiting for the full response. Stream Response toggle * Add `"stream": true` to the custom prompt when streaming is enabled. The saved prompt displays a "streaming" tag. * Enabling streaming disables the "Exit Scenario" field. Streaming applies only to Agent Node and Prompt Node features using OpenAI and Azure OpenAI models. 7. Fill in the **Sample Context Values** and click **Test**. If successful, the LLM response is displayed; otherwise an error appears. Test response 8. **Map the response key:** In the JSON response, double-click the key that holds the relevant information (e.g., `content`). The Platform generates a **Response Path** for that location. Click **Save**. Select response key 9. Click **Lookup Path** to validate the path. Response path 10. Review the **Actual Response** and **Expected Response**: * **Green (match):** Click **Save**. Skip to step 12. Matching responses * **Red (mismatch):** Click **Configure** to open the Post Processor Script editor. Mismatched responses 1. Enter the **Post Processor Script** and click **Save & Test**. Post processor script 2. Verify the result, then click **Save**. The responses turn green. Post processor result 11. (Optional) If **Token Usage Limits** are enabled for your custom model, map the token keys for accurate tracking: * **Request Tokens key:** `usage.input_tokens` * **Response Tokens key:** `usage.output_tokens` Token key mapping Without this mapping, the Platform can't calculate token consumption, which may lead to untracked usage and unexpected costs. 12. Click **Save**. The prompt appears in the Prompts Library.