Skip to content
Get packs

Guide · Agent skill

How to use the Gemini Live API skill: install and build a client

Updated · written with AI from the project's own documentation and checked against our catalogue data

The gemini-live-api skill is an agent skill, a folder of instructions that an AI coding agent loads when it is relevant to your request. With it installed, the agent can scaffold a client service that connects to Google's Gemini Live API over WebSockets for two-way streaming of text, audio and video. This guide explains what the skill does, how to install it, how to trigger it, and what to check before you run the generated code.

Gemini Live API Client Buildergoogle/skills · Apache-2.0

Quick start

  1. 1Download and unzip the skill. Download the gemini-live-api ZIP from this page and unzip it. Place the folder in ~/.claude/skills/ or in your project's .claude/skills/ folder.
  2. 2Authenticate with Google Cloud. Make sure your Google Cloud project has the Vertex AI / Gemini Enterprise Agent Platform APIs enabled, then run gcloud auth application-default login on the host.
  3. 3Make a natural request. Ask the agent to build a Gemini Live API client in your chosen language and name a destination folder such as /tmp/liveapi_out.
  4. 4Review the generated setup. Read the setup script and the commands the agent wants to run. Confirm that it uses a venv and does not touch the system Python.
  5. 5Run the setup script. Run the setup script the agent generated to recreate the environment and install dependencies inside the destination folder.
  6. 6Test with the demo UI. Follow how_to_test_with_ui.md to start the backend and open the frontend. Try text, audio, video and interruptions.

What does the gemini-live-api skill make the agent do?

The skill tells the agent to generate a LiveAPI client service class in your chosen programming language, Python by default. The class opens a WebSocket connection to the Live API endpoint, obtains a bearer token through Application Default Credentials, and exposes async methods to send realtime data, send client content and receive server messages.

The generated code is designed to handle session setup and resumption, refresh the bearer token, and reconnect after a go_away signal or an unexpected disconnect. The skill also asks the agent to build a demo frontend and backend so you can test text, audio, video, transcription and interruption handling in a browser.

Install the skill by downloading the ZIP

Download the skill ZIP from this page and unzip it. For a personal install, place the unzipped folder in ~/.claude/skills/. For a single project, place it in that project's .claude/skills/ folder instead.

If you use the Claude apps, you can instead upload the skill under Settings → Capabilities → Skills.

  • Personal install: ~/.claude/skills/gemini-live-api
  • Project install: .claude/skills/gemini-live-api inside your project

Set up Google Cloud before you start

The skill needs a Google Cloud project with the Vertex AI / Gemini Enterprise Agent Platform APIs enabled. The host that runs the generated client also needs Application Default Credentials. Create them with the command below.

Choose a destination output folder for the generated code, for example /tmp/liveapi_out. The skill writes the code, environment and demo there and should not change your system Python.

  • Google Cloud project with the required APIs enabled
  • gcloud CLI with Application Default Credentials
  • Python with venv support (default language)
  • Network access to the Live API endpoint
gcloud auth application-default login

Trigger the skill with a natural request

You do not need a special command. Describe what you want in plain language and mention Live API, streaming, or a real-time voice or video assistant. The skill's description covers building, scaffolding or integrating a client that connects to the Live API websocket, handling session setup and resumption, bearer token refresh, and sending and receiving ClientMessage and ServerMessage protos.

Requests for one-shot calls such as generateContent, embeddings, image or video generation, or fine-tuning are not what this skill is for. The skill's description points to a separate gemini-api skill for those.

  • Example: Scaffold a Python Gemini Live API client in /tmp/liveapi_out that streams audio and text.
  • Example: Build a real-time voice assistant client with session resumption and a demo web UI.

What files does the skill produce?

The skill copies its reference files into your output folder, creates a venv inside that folder with a setup script, implements the client class, writes a test file, and produces guide files. The reference files include client_server_messages.md, client_server_messages.proto and session_manager.md.

The skill also writes how_to_run.md to document the generated class, and how_to_test_with_ui.md to explain how to start the demo service. Read these after generation, since they reflect the code the agent actually wrote.

  • client_server_messages.md and client_server_messages.proto, copied from references
  • session_manager.md guidance for disconnections and resumption
  • how_to_run.md and how_to_test_with_ui.md
  • A setup script that recreates the isolated environment

Safety notes: scripts and network access

This skill runs scripts and uses the network. The agent must run setup commands on your machine, and the generated code and demo service are not security-audited. Review the commands the agent proposes before you approve them, and do not deploy the demo or client to production without a code review.

The skill's own instructions say the agent should install into a venv inside the destination folder and never use sudo pip install. Check that this is what happens on your machine.

  • Read the generated setup script before running it
  • Confirm the agent is not installing into the system interpreter
  • Treat the demo frontend and backend as a local test tool
  • Review the code before any production use

What people use it for

  • Building a real-time voice assistant that streams microphone audio to Gemini Live and plays back model audio.
  • Adding a live video or screenshot input to a Gemini Live session for camera-based questions.
  • Creating a Python starter client that already includes session resumption and bearer token refresh.
  • Testing transcription and interruption handling in a local demo web app before writing your own UI.

Frequently asked questions

How do I install the gemini-live-api skill?

Download the skill ZIP from this page and unzip it. For a personal install, put the folder in ~/.claude/skills/. For one project, put it in that project's .claude/skills/ folder, or upload it in the Claude apps under Settings → Capabilities → Skills.

What do I need to use the Gemini Live API skill?

You need a Google Cloud project with the Vertex AI / Gemini Enterprise Agent Platform APIs enabled, the gcloud CLI with Application Default Credentials, Python with venv support, and network access to the Live API endpoint. The skill also runs scripts on your machine.

Is the gemini-live-api skill free to use?

The skill is licensed under Apache-2.0, as listed on this page. Google Cloud and Gemini API usage may be billed under your own Google Cloud account, and this page does not state those costs.

Can I use this skill for simple Gemini text generation?

No. The skill is for real-time, bidirectional streaming. For one-shot generateContent calls, embeddings, or image and video generation, the skill's description points to a separate gemini-api skill.

Why does the agent say it cannot find the Live API endpoint or fails to authenticate?

Check that the Vertex AI / Gemini Enterprise Agent Platform APIs are enabled in your project, that you ran gcloud auth application-default login on the same host, and that the machine can reach the wss endpoint for your location. These are the prerequisites the skill lists.

Is the generated code safe for production?

Not without review. The generated code and demo service are not security-audited, and the agent must run setup commands on your machine. Review and test the code before any production deployment.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

Use this skill if you are building a real-time voice or video assistant on the Gemini Live API and want generated, runnable Python with auth, reconnection and a demo already designed in. Skip it for simple one-off Gemini calls, and for production systems unless you are ready to review and test the generated code yourself.

Download the skill