Building a Multimodal Agent with the ADK, Amazon Lightsail, and Gemini Flash Live 3.1

--- title: Building a Multimodal Agent with the ADK, Amazon Lightsail, and Gemini Flash Live 3.1 published: true series: AWS tags: python,geminilive,adk,geminicli --- Leveraging the Google Agent Development Kit (ADK) and the underlying Gemini LLM to build Agentic apps using the Gemini Live API with the Python programming language deployed to Amazon Lightsail.  #### Aren’t There a Billion Python ADK Demos? Yes there are. Python has traditionally been the main coding language for ML and AI tools. The goal of this article is to provide a minimal viable basic working ADK streaming multi-modal agent using the latest Gemini Live Models. #### In the Spirit of Mr. McConaughey’s “alright, alright, alright” So what is different about this lab compared to all the others out there? This is one of the first implementations of the latest Gemini 3.1 Flash Live Model with the Agent Development Kit (ADK). The starting point for the demo was an existing Code lab- which was updated and re-engineered with Gemini CLI. The original Codelab- is here: [Way Back Home - Building an ADK Bi-Directional Streaming Agent | Google Codelabs](https://codelabs.developers.google.com/way-back-home-level-3/instructions#0) #### What Is Python? Python is an interpreted language that allows for rapid development and testing and has deep libraries for working with ML and AI: [Welcome to Python.org](https://www.python.org/) #### Python Version Management One of the downsides of the wide deployment of Python has been managing the language versions across platforms and maintaining a supported version. The **pyenv** tool enables deploying consistent versions of Python: [GitHub - pyenv/pyenv: Simple Python version management](https://github.com/pyenv/pyenv) As of writing — the mainstream python version is 3.13. To validate your current Python: ```console python --version Python 3.13.12 ``` #### Amazon Lightsail [Amazon Lightsail](https://aws.amazon.com/lightsail/) is an easy-to-use virtual private server (VPS) provider and cloud platform designed by AWS for simpler workloads, offering developers pre-configured compute, storage, and networking for a low, predictable monthly price. It is ideal for hosting small websites, simple web apps, or creating development environments. More information is available on the official site here: [Amazon's Simple Cloud Server | Amazon Lightsail](https://aws.amazon.com/free/compute/lightsail/?trk=93c1c080-6a58-41f6-b56e-f352c703feb6&sc_channel=ps&ef_id=CjwKCAjwjtTNBhB0EiwAuswYhjYfDAdZvphotoys8sw1RlOcuvMz1mu6mp0MulOUryHwrqsyynfDEhoCxTEQAvD_BwE:G:s&s_kwcid=AL!4422!3!795794191906!e!!g!!amazon%20lightsail!23527793966!192204323906&gad_campaignid=23527793966&gbraid=0AAAAADjHtp_DJwRcBBdbHLaptszTVpfxR&gclid=CjwKCAjwjtTNBhB0EiwAuswYhjYfDAdZvphotoys8sw1RlOcuvMz1mu6mp0MulOUryHwrqsyynfDEhoCxTEQAvD_BwE) And this is the direct URL to the console: ```plaintext https://lightsail.aws.amazon.com/ls/webapp/home/containers ```  #### Gemini Live Models Gemini Live is a conversational AI feature from Google that enables free-flowing, real-time voice, video, and screen-sharing interactions, allowing you to brainstorm, learn, or problem-solve through natural dialogue. Powered by the **Gemini 3.1 Flash Live model** , it provides low-latency, human-like, and emotionally aware speech in over 200 countries. More details are available here: [Gemini 3.1 Flash Live Preview | Gemini API | Google AI for Developers](https://ai.google.dev/gemini-api/docs/models/gemini-3.1-flash-live-preview) The Gemini Live Models bring unique real-time capabilities than can be used directly from an Agent. A summary of the model is also available here: ```plaintext https://deepmind.google/models/model-cards/gemini-3-1-flash-live/ ```  #### Gemini CLI If not pre-installed you can download the Gemini CLI to interact with the source files and provide real-time assistance: ```console npm install -g @google/gemini-cli ``` #### Testing the Gemini CLI Environment Once you have all the tools and the correct Node.js version in place- you can test the startup of Gemini CLI. You will need to authenticate with a Key or your Google Account: ```console ▝▜▄ Gemini CLI v0.33.1 ▝▜▄ ▗▟▀ Logged in with Google /auth ▝▀ Gemini Code Assist Standard /upgrade no sandbox (see /docs) /model Auto (Gemini 3) | 239.8 MB ``` #### Node Version Management Gemini CLI needs a consistent, up to date version of Node. The **nvm** command can be used to get a standard Node environment: [GitHub - nvm-sh/nvm: Node Version Manager - POSIX-compliant bash script to manage multiple active node.js versions](https://github.com/nvm-sh/nvm) #### Agent Development Kit The [Google Agent Development Kit](https://www.google.com/search?q=Google+Agent+Development+Kit&rlz=1CAIWTJ_enUS1114&oq=what+is+the+adk+google&gs_lcrp=EgZjaHJvbWUyBggAEEUYOTIICAEQABgWGB4yCAgCEAAYFhgeMggIAxAAGBYYHjIICAQQABgWGB4yCAgFEAAYFhgeMggIBhAAGBYYHjIKCAcQABgKGBYYHjINCAgQABiGAxiABBiKBTIKCAkQABiABBiiBNIBCDMxODlqMGo3qAIAsAIA&sourceid=chrome&ie=UTF-8&mstk=AUtExfB5Oo7ZHHcDEHu7aqZiPBA2l1c-QGh5dB7xkkDPIiYcn8O1Imt2IHNR7bzA6JnyDCSDCUGpGWTeBW14namlN_QqzJLLI5-px1BE9jfSxwli6njPDPERjm5pRqNP3uC6HhUKiRcTJ1T8x5LHQrCkVxylw7QWg0N8B4dQDIcWpnVX9Gc&csui=3&ved=2ahUKEwjYu-G8p-uSAxXrv4kEHUbpLo0QgK4QegQIARAB) (ADK) is an open-source, Python-based framework designed to streamline the creation, deployment, and orchestration of sophisticated, multi-agent AI systems. It treats agent development like software engineering, offering modularity, state management, and built-in tools (like Google Search) to build autonomous agents. The ADK can be installed from here: [Agent Development Kit (ADK)](https://google.github.io/adk-docs/) #### Where do I start? The strategy for starting multimodal real time agent development is a incremental step by step approach. First, the basic development environment is setup with the required system variables, and a working Gemini CLI configuration. Then, a minimal ADK Agent is built and tested locally. Next — the entire solution is deployed to Amazon Lightsail. #### Setup the Basic Environment At this point you should have a working Python environment and a working Gemini CLI installation. All of the relevant code examples and documentation is available in GitHub. This repo has a wide variety of samples- but this lab will focus on the ‘gemini31-lightsail’ setup. The next step is to clone the GitHub repository to your local environment: ```console cd ~ git clone https://github.com/xbill9/gemini-cli-aws cd gemini31-lightsail ``` Then run **init.sh** from the cloned directory. The script will attempt to determine your shell environment and set the correct variables: ```console source init.sh ``` If your session times out or you need to re-authenticate- you can run the **set\_env.sh** script to reset your environment variables: ```console source set_env.sh ``` Variables like PROJECT\_ID need to be setup for use in the various build scripts- so the **set\_env** script can be used to reset the environment if you time-out. #### Build the User Interface The front end files provide the user interface: ```console xbill@penguin:~/gemini-cli-aws/gemini31-lightsail$ make frontend cd frontend && npm install && npm run build up to date, audited 219 packages in 577ms 49 packages are looking for funding run `npm fund` for details found 0 vulnerabilities > [email protected] build > vite build vite v7.3.1 building client environment for production... ✓ 33 modules transformed. dist/index.html 0.46 kB │ gzip: 0.29 kB dist/assets/index-xOQlTZZB.css 21.60 kB │ gzip: 4.54 kB dist/assets/index-0hbet2qm.js 214.56 kB │ gzip: 67.44 kB ✓ built in 1.01s ``` #### Test The User Interface The mock server test script allows the interface and Browser settings to be set to allow multimedia — without using any external Model calls or tokens: ```console xbill@penguin:~/gemini-cli-aws/gemini31-lightsail$ make mock . ./mock.sh http://127.0.0.1:8080/ Warning: Frontend build not found at /home/xbill/gemini-cli-aws/gemini31-lightsail/frontend/dist Please run 'npm run build' in the frontend directory. INFO: Started server process [20143] INFO: Waiting for application startup. INFO: Application startup complete. INFO: Uvicorn running on http://0.0.0.0:8080 (Press CTRL+C to quit) ``` The Deployed mock front-end will look similar to:  #### Verify The ADK Installation To verify the setup, run the ADK CLI locally with the biometric\_agent: ```console xbill@penguin:~/gemini-cli-aws/gemini31-lightsail$ make testadk . ./testadk.sh connect to local ADK CLI Log setup complete: /tmp/agents_log/agent.20260406_151152.log To access latest log: tail -F /tmp/agents_log/agent.latest.log /home/xbill/.local/lib/python3.13/site-packages/google/adk/cli/cli.py:204: UserWarning: [EXPERIMENTAL] InMemoryCredentialService: This feature is experimental and may change or be removed in future versions without notice. It may introduce breaking changes at any time. credential_service = InMemoryCredentialService() /home/xbill/.local/lib/python3.13/site-packages/google/adk/auth/credential_service/in_memory_credential_service.py:33: UserWarning: [EXPERIMENTAL] BaseCredentialService: This feature is experimental and may change or be removed in future versions without notice. It may introduce breaking changes at any time. super(). __init__ () Running agent biometric_agent, type exit to exit. [user]: hello [biometric_agent]: Scanner Online. [user]: ``` #### Test The ADK Web Interface This tests the Audio / Video ADK agent interactions: ```console xbill@penguin:~/gemini-cli-aws/gemini31-lightsail$ make adk . ./runadk.sh connect on http://127.0.0.1:8000/ 2026-04-06 15:12:38,201 - INFO - service_factory.py:266 - Using in-memory memory service 2026-04-06 15:12:38,201 - INFO - local_storage.py:84 - Using per-agent session storage rooted at /home/xbill/gemini-cli-aws/gemini31-lightsail/backend/app 2026-04-06 15:12:38,202 - INFO - local_storage.py:110 - Using file artifact service at /home/xbill/gemini-cli-aws/gemini31-lightsail/backend/app/.adk/artifacts /home/xbill/.local/lib/python3.13/site-packages/google/adk/cli/fast_api.py:193: UserWarning: [EXPERIMENTAL] InMemoryCredentialService: This feature is experimental and may change or be removed in future versions without notice. It may introduce breaking changes at any time. credential_service = InMemoryCredentialService() /home/xbill/.local/lib/python3.13/site-packages/google/adk/auth/credential_service/in_memory_credential_service.py:33: UserWarning: [EXPERIMENTAL] BaseCredentialService: This feature is experimental and may change or be removed in future versions without notice. It may introduce breaking changes at any time. super(). __init__ () INFO: Started server process [20329] INFO: Waiting for application startup. +-----------------------------------------------------------------------------+ | ADK Web Server started | | | | For local testing, access at http://0.0.0.0:8000. | +-----------------------------------------------------------------------------+ INFO: Application startup complete. INFO: Uvicorn running on http://0.0.0.0:8000 (Press CTRL+C to quit) ``` Then use the web interface — either on the local interface **127.0.0.1** or the catch-all web interface **0.0.0.0** -depending on your environment:  Special note for Google Cloud Shell Deployments- add a CORS **allow\_origins** configuration exemption to allow the ADK agent to run: ```console adk web --host 0.0.0.0 --allow_origins 'regex:.*' ``` #### Lint and Test the Main Python Code The final step is to build, lint, and test the main Python code. To Lint: ```console xbill@penguin:~/gemini-cli-aws/gemini31-lightsail$ make lint ruff check . All checks passed! ruff format --check . 10 files already formatted cd frontend && npm run lint > [email protected] lint > eslint . ``` To Test: ```console ✦ I'll execute the tests using make test. ╭───────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╮ │ ✓ Shell make test [current working directory /home/xbill/gemini-cli-aws/gemini31-lightsail] (Running project tests with pytest.) │ │ │ │ ... first 4 lines hidden (Ctrl+O to show) ... │ │ configfile: ../../pyproject.toml │ │ plugins: anyio-4.11.0 │ │ collected 8 items │ │ │ │ backend/app/biometric_agent/test_agent.py ..... [62%] │ │ test_live_connection.py . [75%] │ │ test_ws_backend.py . [87%] │ │ test_ws_backend_v2.py . [100%] │ │ │ │ ========================================================= 8 passed in 2.59s ========================================================= │ ╰───────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯ ✦ All 8 tests passed successfully. ``` #### Running Locally The main Python Code can then be run locally: ```console xbill@penguin:~/gemini-cli-aws/gemini31-lightsail$ make run . ./biosync.sh Local URL http://127.0.0.1:8080/ 2026-04-06 15:50:47,335 - INFO - System Config: 2.0 FPS, 10.0s Heartbeat Serving static files from: /home/xbill/gemini-cli-aws/gemini31-lightsail/frontend/dist INFO: Started server process [22807] INFO: Waiting for application startup. INFO: Application startup complete. INFO: Uvicorn running on http://0.0.0.0:8080 (Press CTRL+C to quit) ``` Then connect to the local front end:  #### Deploying to Lightsail A utility script runs the deployment to AWS Lightsail. Use the deploy version from the local system: ```console xbill@penguin:~/gemini-cli-aws/gemini31-lightsail$ make deploy ./save-aws-creds.sh Exporting AWS credentials... Successfully saved credentials to .aws_creds The Makefile will now automatically use these for deployments. ./deploy-lightsail.sh 0.0s 0.0s ``` You can validate the final result by checking the messages: ```plaintext Deployment complete. ``` Once the container is deployed- you can then get the endpoint: ```console xbill@penguin:~/gemini-cli-aws/gemini31-lightsail$ make status ------------------------------------- | GetContainerServices | +-------------+-------------+-------+ | Deployment | State | URL | +-------------+-------------+-------+ | ACTIVE | DEPLOYING | None | +-------------+-------------+-------+ xbill@penguin:~/gemini-cli-aws/gemini31-lightsail$ make endpoint https://biometric-scout-service.6wpv8vensby5c.us-east-1.cs.amazonlightsail.com/ xbill@penguin:~/gemini-cli-aws/gemini31-lightsail$ ``` The service will be visible in the Lightsail console:  #### Running the Web Interface Start a connection to the Lightsail deployed app: ```plaintext https://biometric-scout-service.6wpv8vensby5c.us-east-1.cs.amazonlightsail.com/ ``` Then connect to the app :  Then use the Live model to process audio and video:  Finally — complete the sequence:  #### So What Really Changed in the Lab? It looks the Same to Me! This project was a great learning experience to get first hand review of working with different LLM models and common issues with ADK and the nuances of working with a real-time Live model. The key changes include: - Gemini Models vs Vertex AI Models The original lab was using LLM models hosted on Vertex AI with PROJECT\_ID and REGION authentication. The Gemini 3.1 Live Flash model is only current supported with the Gemini API- which requires an API key. When the authentication method and API is changed- the actual model support also changes — so you have to be careful that the model you want to use is available on that API. - Support for Gemini 3.1 Live The API support for the 3.1 version of the Live model is slightly different- and required a translation layer for the current version of the ADK. This “monkey patch” acts as a workaround until full support of the 3.1 Live model is available in the ADK. There are several open issues in GitHub around these issues and they are noted in the **GEMINI.md** file. - Cloud Shell CORS over-ride The recent versions of the ADK are more restrictive for running from the Cloud Shell. - ADK CLI Support The ADK agent was extended to detect when running from the CLI and not throw errors related to the Live model. - Audio Support Client Side The client side audio API was updated to the recommended Worklet implement ion to get audio processing off the main thread. - Re-Architected the Protocol to be Mostly JSON-Free The original version wrapped the binary messages in JSON- which makes it easier to work with- but also introduces overhead. The updated code sends the Audio and Video streams as raw data. - Gemini CLI Development support **README.md** and **GEMINI.md** were added to the project- as well as an Agent Skill for working with Live Models. #### Gemini CLI Code Review As a final step — Gemini CLI was used for a full code review of the project: ```plaintext ✦ Based on my comprehensive review of the "Alpha Rescue Drone - Biometric Security System," the codebase is exceptionally well-structured, follows modern engineering standards, and strictly adheres to the specialized requirements for Gemini 3.1 Flash Live and the Google Agent Development Kit (ADK). Core Architecture & Integration The system excels at orchestrating a high-frequency, multimodal data stream. The use of a binary protocol (0x01 for audio, 0x02 for video) in useGeminiSocket.js minimizes overhead, while the monkey-patching utility in patch_adk.py demonstrates a sophisticated understanding of the evolving Gemini 3.1 Live API. This proactive approach ensures compatibility with features like send_realtime_input that are not yet natively handled by the standard ADK. Backend Analysis - Agent Intelligence: The instructions in agent.py are precise and prioritize safety (offensive gesture detection) and hidden overrides ("Devil's Horns") effectively. - Resilience: The backend implements a heartbeat stimulus (CONTINUE_SURVEILLANCE) and a neural handshake, which are critical for maintaining model engagement during visual-only periods. - Model Fallback: The get_model_id() logic is clever, automatically switching to gemini-2.5-flash for CLI runs to prevent 404 errors, while defaulting to the high-performance gemini-3.1-flash-live-preview for production. Frontend & UI/UX - Aesthetic: The "Neon Cyan" HUD in BiometricLock.jsx provides a high-fidelity, futuristic feel that aligns with the "Mission Alpha" theme. - Feedback Loop: The synchronization between the random sequence generation, visual matching, and the special "Heavy Metal" audio event creates a compelling user experience. - Efficiency: Utilizing toBlob with JPEG compression (0.6) for the 2 FPS video stream is an optimal choice for balancing visual clarity with network latency. Testing & Maintainability - Validation: The project maintains a clean slate with 100% test and linting compliance. The inclusion of mock.sh is a significant developer productivity boost. - Standards: The use of Ruff for Python and ESLint for the frontend ensures long-term code quality. Recommendations 1. Connection Resilience: While the SYSTEM_ERROR closes the socket correctly, the frontend could benefit from a more graceful "Reconnecting..." state if the network fluctuates during a mission. 2. Patch Management: As the Gemini Live API stabilizes, keep an eye on official ADK updates to deprecate the manual unrolling of media_chunks in patch_adk.py. 3. Security: Verify that the .env variables are correctly injected into the Cloud Run environment via the Makefile or cloudbuild.yaml without being exposed in the build logs. Overall, the system is a robust and creative implementation of real-time multimodal AI. Authentication Protocol: Stable. 🤘 ? for shortcuts ``` #### Summary The Agent Development Kit was used to enable a multi-modal agent using the Gemini Live Model. This Agent was tested locally with the CLI and then deployed to Amazon Lightsail. Several key take-aways and lessons learned were summarized from working with the transition to a new Live Gemini LLM model. Finally, Gemini CLI was used for a complete project code review.