# Agora ESP32 Large Model Intelligent Dialogue *[简体中文](./README.cn.md) | English* ## Overview This is an RTC Client SDK & Demo running on Espressif ESP32-S3 Korvo V3 development board. This example demonstrates how to make TEN-Agent work with it. ### File Structure ``` ├── CMakeLists.txt ├── components Agora IoT SDK component │   ├── agora_iot_sdk │   │   ├── CMakeLists.txt │   │   ├── include Agora IoT SDK header files │   │   │   ├── agora_rtc_api.h │   │   └── libs Agora IoT SDK libraries │   │   ├── libagora-cjson.a │   │   ├── libahpl.a │   │   ├── librtsa.a ├── main LLM Demo code │   ├── ai_agent.h │   ├── app_config.h │   ├── common.h │   ├── audio_proc.h │   ├── rtc_proc.h │   ├── CMakeLists.txt │   ├── Kconfig.projbuild | ├── ai_agent.c | ├── audio_proc.c | ├── rtc_proc.c │   └── llm_main.c ├── partitions.csv Partition table ├── README.en.md ├── README.md ├── sdkconfig.defaults └── sdkconfig.defaults.esp32s3 ``` ## Environment Setup ### Hardware Requirements This example currently supports the `ESP32-S3-Korvo-2 V3` development board only. ## Compilation and Download ### Agora IOT SDK To compile and run this example, you need the Agora IoT SDK. The SDK can be downloaded at [here](https://rte-store.s3.amazonaws.com/agora_iot_sdk.tar) Put `agora_iot_sdk.tar` to `esp32-client/components` directory and run the following command: ```bash cd esp32-client/components tar -xvf agora_iot_sdk.tar ``` ### Linux Operating System #### Default IDF Branch This example supports IDF tag v[5.2.3] and later versions, with the default set to IDF tag v[5.2.3] (commit id: c9763f62dd00c887a1a8fafe388db868a7e44069). To select the correct IDF branch, run the following commands: ```bash cd $IDF_PATH git checkout v5.2.3 git pull git submodule update --init --recursive ``` This example supports ADF v2.7 tag (commit id: 9cf556de500019bb79f3bb84c821fda37668c052). #### Applying the IDF Patch A patch must be applied to IDF. Use the following command: ```bash export ADF_PATH=~/esp/esp-adf cd $IDF_PATH git apply $ADF_PATH/idf_patches/idf_v5.2_freertos.patch ``` #### Compiling the Firmware Copy the example project directory (esp32-client) to the `~/esp` directory and run the following commands: ```bash $ . $HOME/esp/esp-idf/export.sh $ cd ~/esp/esp32-client $ idf.py set-target esp32s3 $ idf.py menuconfig --> Agora Demo for ESP32 --> (Configure WIFI SSID and Password) $ idf.py build ``` To configure FreeRTOS backward compatibility: In `menuconfig`, navigate to `Component config` --> `FreeRTOS` --> `Kernel` and enable `configENABLE_BACKWARD_COMPATIBILITY`. ### Windows Operating System #### Default IDF Branch Download IDF, selecting version v5.2.3 (offline version) from the following link: [ESP-IDF Windows Setup](https://docs.espressif.com/projects/esp-idf/zh_CN/v5.2.3/esp32/get-started/windows-setup.html) Download ADF to the `Espressif/frameworks` directory to support ADF v2.7 tag (commit id: 9cf556de500019bb79f3bb84c821fda37668c052): [ESP-ADF Setup](https://docs.espressif.com/projects/esp-adf/zh_CN/latest/get-started/index.html#step-2-get-esp-adf) #### Applying the IDF Patch Method 1: Add `ADF_PATH` to the environment variables in system settings: ``` E:\esp32s3\Espressif\frameworks\esp-adf ``` Method 2: Add `ADF_PATH` via the command line: ```bash $ setx ADF_PATH Espressif/frameworks/esp-adf ``` **Note:** After setting the `ADF_PATH` environment variable, restart ESP-IDF 5.2 PowerShell for changes to take effect. Apply the required patch to IDF using: ```bash cd $IDF_PATH git apply $ADF_PATH/idf_patches/idf_v5.2_freertos.patch ``` #### Compiling the Firmware Copy the example project directory (esp32-client) to the `Espressif/frameworks` directory and run the following commands: ```bash $ cd ../esp32-client $ idf.py set-target esp32s3 $ idf.py menuconfig --> Agora Demo for ESP32 --> (Configure WIFI SSID and Password) $ idf.py build ``` Configure FreeRTOS backward compatibility: In `menuconfig`, navigate to `Component config` --> `FreeRTOS` --> `Kernel` and enable `configENABLE_BACKWARD_COMPATIBILITY`. ### Flashing the Firmware Run the following command: ```bash $ idf.py -p /dev/ttyUSB0 flash monitor ``` **Note:** On Linux, you might encounter permission issues with `/dev/ttyUSB0`. Run the following command to fix it: ```bash sudo usermod -aG dialout $USER ``` Once flashing is complete, the example will run automatically. After the device joins the RTC channel, the serial output will display: **"Agora: Press [SET] key to Join the Ai Agent ..."** ## How to Use the Example ### Quick Start in 5 Minutes **Note:** Ensure at least one speaker is connected to the development board. ### Configure your own agent Let's walk you through creating your own user and device accounts. #### Update app_config.h 1. Open the `app_config.h` file in the project. 2. Modify `TENAI_AGENT_URL` to the URL of your TEN-Agent server (usually the 8080 port service). 3. Modify `AI_AGENT_CHANNEL_NAME` to a unique channel name. 4. If you have your Ten-Agent configured for `openai_v2v` graph working, you should be good to go. 5. If you want to use other graphs, you can modify how request is constructed in `ai_agent.c` file the `_build_start_json` method. 6. Recompile and flash the firmware. #### Demo: Real-time Voice Dialogue with Large Model AiAgent 1. Press the `SET` button to start the large model. 2. Press the `MUTE` button to stop the large model. 3. Press the `VOL+` button to increase volume (increments of 10, up to a maximum of 100). 4. Press the `VOL-` button to decrease volume (decrements of 10, down to a minimum of 0). 5. After the device boots up, it will automatically connect to the RTC channel associated with the generated APPID. Press the `SET` button to initiate real-time voice dialogue; press the `MUTE` button to stop it. ## About Agora Agora’s audio and video IoT platform leverages its proprietary real-time transmission network, **Agora SD-RTN™ (Software Defined Real-time Network)**, to provide real-time audio and video streaming capabilities for Linux/RTOS devices with networking capabilities. The solution ensures high connectivity, real-time performance, and stability even under uncertain network conditions using advanced techniques such as forward error correction, intelligent retransmission, bandwidth prediction, and stream smoothing. Additionally, it offers a minimal memory footprint, making it ideal for resource-constrained IoT devices, including the entire Espressif ESP32 series. ## Technical Support For technical support, follow the links below: - Report bugs and inquiries directly to the community representatives. We will respond as soon as possible.