ByteGPT | Templates v1 and v2

Introduction

The new updates of ByteGPT present the templates v2, which facilitate configuration, expand support for different types of media (voice, image, file and sticker) and open space for advanced integrations — such as sending data in JSON to Dify. If you are on v1 and thinking about migrating, or are choosing the best version to start with, this guide shows the main differences, advantages and best practices.

O ByteGPT can operate with ChatGPT (API) or with various “Providers” (OpenAI, Dify, n8n, chatbase). Each offers a unique way of integrate AI capabilities to its flow in Kommo CRM — and by connecting specifically to the Dify, you orchestrate virtually any language model (OpenAI, Gemini, Claude, DeepSeek, among others) in one place, customizing chatbot flows, sending HTTP requests and using advanced automation tools.

1. Overview of ByteGPT

O ByteGPT is a solution of Artificial Intelligence integrated into Kommo CRM , capable of creating a hybrid chatbot that combines traditional Salesbot flows with advanced language models (GPT). In practice, this makes it possible to:

• Answer frequently asked questions 24/7.
• Interpret messages in natural language (including voice and image).
• Offer intelligent and personalized service, as the ByteGPT can be configured with your company's specific data and prompts.

With the evolution towards the Version 2 (v2) of the templates, the ByteGPT has become even more flexible and easy to maintain, especially in terms of media handling and integration with platforms such as Dify and n8n.

2. Why are there two versions of templates (v1 and v2)?

Version v1

• It was the first generation of templates ByteGPT, initially focused on text and audio.
• The audio transcription is done after passing through the ChatGPT Webhook (step 187) and then sent as text (transcribed) to the Assistant.
• allows the image interpretation, but depends on specific settings in the stream, according to the field return ByteGPT+ | LOG, triggering a specific prompt in step 235 if the field ByteGPT+ | LOG contains #image. This allows you to customize the analysis or request as needed.
• It remains functional and serves those who do not need more advanced media identification features or “cleaner” streams.

Version v2

• Offers reliable, more organized templates, with the option of enable/disable the identification of an audio message, image, file or sticker.
• Audio transcription is done automatically and sent in text format (transcribed) for the Assistant to respond to. You don't need to go through the Webhook to transcribe.
• It facilitates the development of customized flows within Salesbot, as each type of media (#voice, #picture, #sticker, #file) is automatically detected in the LOG, allowing specific branches.
• It integrates in a way more transparent to Dify and n8n, enabling data sending in JSON, audio transcription and image analysis.
• Exige suitability for those using the v1 template (e.g. removing “LOG OK” check from the template), but brings greater flexibility and clarity to the Salesbot blocks in the v2 template.

In summary: v2 is an evolution that makes the configuration of the ByteGPT more intuitive, but v1 remains available for those who are already used to it or do not need the new features.

3. Main changes and new features in Version 2

3.1. Cleaner and more modular templates

v2 flows have been reworked to facilitate reading and maintenance. Salesbot blocks are more segmented and organized, allowing you to adjust prompts and responses intuitively, making the cleaner flow and eliminating redundant nodes.

3.2. Native identification of message types

In v2, audio, image, file and sticker identification is done internally by our API, without the need for a specific webhook in Salesbot to recognize the media type. For this, there is a 5 second interval in Salesbot block, the time it takes for the API to parse the message and return the appropriate LOG — for example, #voice, #picture, #sticker, or #file. This way, Salesbot can direct each type of content to the correct block.

Already in v1, a generic “OK” LOG was used to confirm actions, which limited the creation of specific branches for each media format.

3.3. Processing of voice, images, files and stickers

The v2 formalizes the detection and treatment of each type of media, without having to extend the flow. This expands the field of action of ByteGPT to:

• Voice recognition: automatically transcribing audio.
• Image interpretation: with targeted prompts (e.g., analyzing the content of a photo).
• Files and stickers: being able to send them to the Assistant (OpenAI, Dify, n8n or Chatbase) or treat them according to the logic of your flow in Kommo.

3.4. Updated V2 Handler and pauses

• Anyone migrating from v1 to v2 will need to update the handler URL by adding “V2” at the end (e.g. …/handlerV2).
• It is recommended to insert a small “pause”, of 5 seconds (for example, in step 253 of the template), to give time to ByteGPT to identify the type of media before responding.

3.5. Advanced integration with Dify or n8n

v2 makes it even clearer integration with Dify or n8n:

• Audios are transcribed automatically, without requiring any extra configuration.
• In the case of images and files, if Dify or n8n is able to receive them, the ByteGPT can generate specialized prompts to analyze this content.
• the flow of sending in JSON allows you to pass field values ​​from Kommo for Dify or n8n to execute custom logic.

4. New Templates in v2: The Three Main Variations

When accessing the new templates ByteGPT v2, you will find (usually) three business:

1. ChatGPT + Provider (OpenAI, Dify, n8n, Chatbase)

• Combination of ChatGPT and a Provider (OpenAI, Dify, n8n, or Chatbase) in the same flow.
• ChatGPT can be used for internal tasks — for example, an “intent prompt” to analyze the message without involving the main Assistant, or to generate examples, summaries, and point-in-time analysis.
• Parallel usage avoids “contamination” in the context of the main Assistant, keeping company-specific instructions intact.
• It's possible activate or deactivate the Assistant as needed, or even combine both — for example, ChatGPT assesses intent and then redirects the conversation to the primary Assistant, which has the instructions and business knowledge.

2. JSON to Dify or n8n

• Enables send data in JSON format directly to Dify or n8n, bringing flexibility to integrate specific fields of Kommo (e.g. leads or contacts).
• Instead of passing this information in the user message (sys.query), you send it as input parameters for Dify or n8n. You must create input variables (in the Start block) in Dify or n8n. The same name as the Dify variable must be entered in the Salesbot block. Kommo, and inform the field of Kommo corresponding.
• Allows you to search for external data, create conditions and insert variables in Dify or n8n instructions, enabling advanced flows of automation and custom logic.

Template difference “JSON to Dify or n8n”

The big difference is the possibility of set JSON value which will be sent to Dify or n8n. For example, entering something like:

{
  "variavel_dify1":"[Campo Kommo 1]",
  "variavel_dify2":"[Campo Kommo 2]",
  "variavel_dify3":"[Campo Kommo 3]"
}

The last line has no comma.

Make sure that each variable defined in “dify_variable1” is also created within your Dify application.

You can enter as many fields as you want, separating them with commas (except the last one). Pay attention to the extra spaces, which should not exist.

Learn more about field updates in Kommo with Dify or n8n in the article: How to update fields Kommo using ByteGPT + Dify or n8n

3. Provider (OpenAI, Dify, n8n, Chatbase)

• In this option, only the Provider is used — all interaction and prompt customization occurs within the chosen Provider itself (OpenAI, Dify, n8n or Chatbase), PULL involve ChatGPT in parallel steps.
• It is a more model lean and direct, ideal for those who want all discussions to remain within the Assistant, with greater freedom to customize the flow.
• In the case of Dify, it is still possible to use the question classifier (acting as an intent prompt) or configure specific routes for different scenarios, providing maximum flexibility in the construction of the flow.

Template for n8n:

5. How to choose between v1 and v2?

Scenario 1: Basic flows and no demand for new media

• If your flows only work with simple text and audio, without needing to differentiate or analyze images, files or stickers, and do not require JSON integration with Dify or n8n, v1 works well.
• If everything is running smoothly and there is no need for more complex features, you can keep v1 without any worries.

Scenario 2: Need for voice, image, file and advanced integrations

• If you want treat different types of media (voice, image, files, stickers) natively in the flow, v2 does this internally, without relying on LOG “OK” or extra configurations.
• For those who wish dynamic integrations with Dify or n8n (like sending CRM data via JSON), create specific branches and take advantage of stability improvements, v2 is the best choice.

Scenario 3: Starting now or planning to grow

• If you is starting to use the ByteGPT, or foresees implement more features in the future, v2 is strongly recommended.
• It brings greater organization, compatibility with future features and facilitates expansion for new media, integrations and advanced workflows as needed.

6. Integration with Dify and sending media

For Dify to accept media (images, stickers and files) sent from Kommo by the ByteGPT, follow these steps:

1. In the Dify flow, open Resources and enable File Upload.

2. Select the types of media you want to receive (images, documents, etc.).

3. Enable the Vision option, add the sys.files variable to the Vision section, and choose a Vision-compatible LLM model (e.g. GPT-4.1 or similar).

This way, every file received will automatically arrive at the field sys.files. Then add a block IF/ELSE that use sys.files to direct each type of message to the appropriate treatment.

Observation: for messages from voice, ByteGPT already transcribes the audio automatically and sends the text directly to Dify, without requiring any additional configuration.

Best practices when using Dify

1. Select the appropriate LLM model:
   • If you want to analyze images (Vision) or handle files, activate or choose a model that supports these functions within Dify.
   • In case of sending images, check if the model recognizes or interprets images, otherwise, use a specific block that supports visual content analysis.
2. Document Extraction Blocks:
   • To manipulate text or PDF files, use the extraction block inside Dify.
   • This block allows you to “read” the file contents and forward them to another LLM block for further analysis or summary.
3. Reference to sys.files in LLM blocks:
   • When sending files (image, PDF, etc.) via ByteGPT, Dify can record this content in the sys.files field.
   • Make sure indicate sys.files in the LLM block so that the model has access to the file and can interpret it correctly.
4. Using JSON for specific variables:
   • If you want to pass on information about Kommo (lead ID, contact data, etc.) or call external APIs, take advantage of integration via JSON.
   • You can parameterize variables in the Salesbot flow and send them to Dify, which will execute the necessary logic or query before returning the response.
5. Application Examples:
   • Audios: O ByteGPT sends the audio to Dify, which automatically transcribes it and returns the text for processing in the stream.
   • Images: If configured to accept images, Dify interprets the file and returns a description or analysis as per your prompt.
   • Documents: Use extraction blocks to process file content, then route the text to an LLM to generate a summary, insights, or contextualized answers.

By following these recommendations, you will leverage the full potential of Dify to create smarter, more integrated and personalized chatbot flows in ByteGPT.

7. Migration: What to consider when switching to v2

• Handler: change the URL, adding “V2” (e.g.: …/handlerV2).
• LOG “OK” check: if your old flow depended on this, review it so it doesn't cause errors (v2 doesn't use this LOG).
• Pause for media identification: include a range (for example, in step 253) for the ByteGPT properly process the received message.
• Testing in Salesbot: Simulate sending audio messages, images, files, etc., to ensure that each branch of the flow is working.

8. Final considerations

A 2 Version of the templates of the ByteGPT represents a big leap for those who want to automate service via text, voice, image or format files organized, efficient and scalable. Your flows are cleaner and modular configurations make it simple to create different experiences for each type of media, in addition to advanced integrations (such as JSON + Dify).

• If you already uses v1 and want to migrate, just review the flow to remove the dependency on “LOG OK”, update the handler to “handlerV2” and adjust the pauses or checks required for each type of media.
• Se will start from scratch, choose v2 directly to enjoy all the latest features.
• You can use MCP (Model Context Protocol) easily through n8n and Dify, or even interconnect the two (Dify calling some n8n MCP, for example), facilitating integrations between tools such as Google Calendar, Google Sheets, Database and many others.

Recommendation: Use Dify or n8n to enhance your AI project

O Dify is an excellent way to orchestrate different AI models (OpenAI, Gemini, Claude, among others), as it offers greater flexibility and automation — whether in a subscription to the Dify plans or in the self-hosted. To understand the advantages of using Dify instead of just OpenAI, we recommend:

• Why use Dify in a chatbot project?
• Video: Dify + ByteGPT + Kommo: The power of AI for more efficient CRM

Next Steps

If you have any questions about integrations, sending media ou special prompts, check out our tutorials and additional materials from Bytebio or contact support.

Regardless of the scenario — technical support, pre-sales, FAQ ou multimedia streams — v2 is the ideal foundation for harnessing the full potential of AI in Kommo CRM. If you need help with setup or migration, talk to technical support. The team Bytebio is ready to guide and ensure that your ByteGPT reach the maximum performance.