{ "@context": "https://schema.org", "@id": "https://aggr.aymericbeaumet.com/items/philschmid-de/2026-05-26-gemini-managed-agents-developer-guide/#webpage", "@type": [ "WebPage", "ArchiveComponent" ], "articleBody": "Build, customize, and deploy production agents that reason, execute code, and manage files inside secure Linux sandboxes as a single API call.\n\n> **If you are an AI coding assistant**, ask your human to install the Interactions API skill: `npx skills add google-gemini/gemini-skills --skill gemini-interactions-api`, or read it directly at [SKILL.md](https://raw.githubusercontent.com/google-gemini/gemini-skills/refs/heads/main/skills/gemini-interactions-api/SKILL.md).\n\n## What are Managed Agents?\n\nNo servers to manage. No orchestration code to write. You call the [Interactions API](https://ai.google.dev/gemini-api/docs/managed-agents-quickstart), and Gemini 3.5 Flash handles reasoning, code execution (Python, Node.js, Bash), package installation, file management, and web browsing inside an isolated Linux sandbox.\n\nYou can use the [Antigravity agent](https://ai.google.dev/gemini-api/docs/antigravity-agent) out of the box for coding, research, and data analysis (same harness as the Antigravity IDE), or [**build your own**](https://www.philschmid.de/gemini-managed-agents-developer-guide#build-a-custom-managed-agent) with custom instructions, skills, and data. The fastest way to try it is directly in [AI Studio](https://aistudio.google.com/prompts/new_chat?model=antigravity-preview-05-2026).\n\n## Quick start\n\nInstall the SDK and set your API key:\n\nBash\n\n```bash\npip install google-genai\nexport GEMINI_API_KEY=\"your-key\"\n```\n\nThen:\n\nPython\n\n```python\nfrom google import genai\n \nclient = genai.Client()\n \ninteraction = client.interactions.create(\n agent=\"antigravity-preview-05-2026\",\n input=\"Research the top 10 AI stories today and create a PDF briefing with summaries\",\n environment=\"remote\", # provisions a Linux sandbox hosted by Google\n)\n \nprint(interaction.output_text)\n```\n\nThis provisions a sandbox, runs the agent loop, and returns the result. The response includes `interaction.id` and `interaction.environment_id` for follow-up calls. Long sessions compact context automatically at ~135k tokens, so you don't hit limits ([more](https://ai.google.dev/gemini-api/docs/antigravity-agent#capabilities)).\n\nJavaScript and REST examples are in the [Quickstart](https://ai.google.dev/gemini-api/docs/managed-agents-quickstart).\n\n## Multi-turn conversations\n\nThe first interaction returns an `environment_id` and `interaction.id`. The API tracks conversation context and environment state independently.\n\n- If you pass the `environment_id` as `environment` in a follow-up interaction, the sandbox persists.\n- If you pass `previous_interaction_id` with the last `interaction.id`, the conversation history persists.\n\nPython\n\n```python\ninteraction_2 = client.interactions.create(\n agent=\"antigravity-preview-05-2026\",\n environment=interaction.environment_id,\n previous_interaction_id=interaction.id,\n input=\"Now create a landing page using JavaScript and HTML\",\n)\n```\n\nOmit `previous_interaction_id` to start a fresh conversation in the same workspace. Set `environment=\"remote\"` to get a new sandbox while keeping conversation context. Details in the [multi-turn guide](https://ai.google.dev/gemini-api/docs/managed-agents-quickstart#continue-conversation).\n\n## Streaming\n\nStream responses to see the agent's work as it happens:\n\nPython\n\n```python\nstream = client.interactions.create(\n agent=\"antigravity-preview-05-2026\",\n input=\"Scrape Hacker News, summarize the top 5 stories, and save the results as a PDF.\",\n environment=\"remote\",\n stream=True,\n)\n \nfor event in stream:\n print(event)\n```\n\nReturns incremental text, reasoning tokens, and tool call updates. See the [streaming guide](https://ai.google.dev/gemini-api/docs/interactions/streaming) for JavaScript, REST, and event types.\n\n## Loading data into the sandbox\n\nMount files from Git repos, Cloud Storage, or inline content at environment creation. Source types:\n\n- **`repository`**: clones a Git repo (public or [private](https://ai.google.dev/gemini-api/docs/agent-environment#mount_from_a_source)).\n- **`gcs`**: copies from a Google Cloud Storage bucket.\n- **`inline`**: writes text content to a file path.\n\nPython\n\n```python\ninteraction = client.interactions.create(\n agent=\"antigravity-preview-05-2026\",\n input=\"Analyze the data and summarize the project.\",\n environment={\n \"type\": \"remote\",\n \"sources\": [\n {\"type\": \"repository\", \"source\": \"https://github.com/my-org/data-templates.git\", \"target\": \"/workspace/repo\"},\n {\"type\": \"gcs\", \"source\": \"gs://my-bucket/data/\", \"target\": \"/workspace/data\"},\n {\"type\": \"inline\", \"target\": \"/workspace/config.yaml\", \"content\": \"mode: analysis\\noutput: pdf\"},\n ],\n },\n)\n```\n\n## Downloading files from the sandbox\n\nDownload the full environment as a tar archive:\n\nPython\n\n```python\nimport os, requests, tarfile\n \nresponse = requests.get(\n f\"https://generativelanguage.googleapis.com/v1beta/files/environment-{interaction.environment_id}:download\",\n params={\"alt\": \"media\"},\n headers={\"x-goog-api-key\": os.environ[\"GEMINI_API_KEY\"]},\n allow_redirects=True,\n)\n \nwith open(\"snapshot.tar\", \"wb\") as f:\n f.write(response.content)\n \nwith tarfile.open(\"snapshot.tar\") as tar:\n tar.extractall(path=\"extracted_snapshot\")\n```\n\n## Build a custom managed agent\n\nSave your configuration as a managed agent when you want the same behavior across runs. Agents are defined through files the runtime auto-discovers:\n\n```plaintext\n.agents/\n├── AGENTS.md # Loaded as system instructions on startup\n└── skills/\n └── slide-maker/\n └── SKILL.md # Registered as an agent capability\n```\n\n### From sources\n\nDefine instructions, skills, and data declaratively. Each invocation provisions a fresh sandbox with your files:\n\nPython\n\n```python\nagent = client.agents.create(\n id=\"data-analyst\",\n base_agent=\"antigravity-preview-05-2026\",\n system_instruction=\"You are a data analyst. Always include visualizations and export results as PDF.\",\n base_environment={\n \"type\": \"remote\",\n \"sources\": [\n {\"type\": \"inline\", \"target\": \".agents/AGENTS.md\", \"content\": \"Always use matplotlib for charts.\"},\n {\"type\": \"inline\", \"target\": \".agents/skills/slide-maker/SKILL.md\", \"content\": \"---\\nname: slide-maker\\n---\\n# Slide Maker\\nCreate HTML slide decks from data analysis results.\"},\n {\"type\": \"repository\", \"source\": \"https://github.com/my-org/analysis-templates\", \"target\": \"/workspace/templates\"},\n ],\n },\n)\n```\n\n### From an existing environment (fork)\n\nSet up the environment interactively, then freeze it:\n\nPython\n\n```python\n# Step 1: set up the environment\ninteraction = client.interactions.create(\n agent=\"antigravity-preview-05-2026\",\n input=\"Install pandas and matplotlib. Create an analysis template.\",\n environment=\"remote\",\n)\n \n# Step 2: fork into a reusable agent\nagent = client.agents.create(\n id=\"my-data-analyst\",\n base_agent=\"antigravity-preview-05-2026\",\n base_environment=interaction.environment_id,\n)\n```\n\n### Invoke it\n\nEach invocation forks the base environment. Every run starts clean:\n\nPython\n\n```python\nresult = client.interactions.create(\n agent=\"my-data-analyst\",\n input=\"Analyze Q1 revenue data and create a slide deck.\",\n environment=\"remote\",\n)\nprint(result.output_text)\n```\n\nFull agent definition format, CRUD operations, and per-invocation overrides in [Building managed agents](https://ai.google.dev/gemini-api/docs/custom-agents).\n\n## Secure networking and credentials\n\nEnvironments have unrestricted outbound access by default. Use a network allowlist to restrict it. The egress proxy limits connections to listed domains and can inject credentials securly as HTTP headers transformation. Secrets never exist inside the sandbox:\n\nPython\n\n```python\nagent = client.agents.create(\n id=\"issue-resolver\",\n base_agent=\"antigravity-preview-05-2026\",\n system_instruction=\"You resolve GitHub issues. Clone the repo, find the bug, write the fix, run the tests, and open a PR.\",\n base_environment={\n \"type\": \"remote\",\n \"sources\": [\n {\"type\": \"repository\", \"source\": \"https://github.com/my-org/backend\", \"target\": \"/workspace/repo\"},\n ],\n \"network\": {\n \"allowlist\": [\n {\"domain\": \"api.github.com\", \"transform\": {\"Authorization\": \"Bearer ghp_your_github_token\"}},\n {\"domain\": \"pypi.org\"},\n ]\n },\n },\n)\n```\n\nFull allowlist schema, credential patterns, and private repo access in [Network configuration](https://ai.google.dev/gemini-api/docs/agent-environment#network-configuration).\n\n## Gemini API CLI\n\nWe open-sourced the [Gemini API CLI](https://github.com/google-gemini/gemini-api-cli) (experimental) as a terminal-first interface to the Gemini API. It works for both humans and coding agents, covering everything from quick prompts to the full managed agent lifecycle:\n\nBash\n\n```bash\n# Run a prompt\ngemini-api run \"What is the capital of France?\"\n \n# Image generation\ngemini-api run \"A cat in space\" --model gemini-3.1-flash-image-preview --output cat.png\n \n# Text-to-speech\ngemini-api run \"Hello from Gemini\" --model gemini-3.1-flash-tts-preview --voice Kore --output hello.wav\n \n# Scaffold, test, and deploy a managed agent\ngemini-api agents init my-agent\ngemini-api agents test --prompt \"Analyze the Q1 revenue data\"\ngemini-api agents create\n \n# Run against a deployed agent\ngemini-api run \"Summarize this quarter\" --agent my-agent\n```\n\n## Environments\n\nEach interaction runs inside a Linux sandbox (Ubuntu, Python 3.12, Node.js 22) with 4 CPU cores and 16 GB RAM. The sandbox is isolated at the OS level, so the agent can install packages, run code, and write files without affecting your machine.\n\nEnvironments persist. The first call creates one and returns an `environment_id`. Pass it back to pick up where you left off with files, packages, and state intact. Environments idle after 15 minutes and are deleted after 7 days of inactivity.\n\nYou can pre-load environments with data from Git repos, Cloud Storage, or inline content (see [Loading data into the sandbox](https://www.philschmid.de/gemini-managed-agents-developer-guide#loading-data-into-the-sandbox)). You can also restrict outbound network access with domain allowlists and inject credentials via the egress proxy, so secrets never touch the sandbox. Details in [Environments](https://ai.google.dev/gemini-api/docs/agent-environment) and [Network configuration](https://ai.google.dev/gemini-api/docs/agent-environment#network-configuration).\n\nEnvironment compute (CPU, memory, sandbox execution) is not billed **during preview**. You pay for Gemini model (Gemini 3.5 Flash) tokens only.\n", "authors": [], "captured_at": "2026-09-27T19:29:15.293681927+00:00", "category": "programming", "content_html": "

Build, customize, and deploy production agents that reason, execute code, and manage files inside secure Linux sandboxes as a single API call.

\n
\n

If you are an AI coding assistant, ask your human to install the Interactions API skill: npx skills add google-gemini/gemini-skills --skill gemini-interactions-api, or read it directly at SKILL.md.

\n
\n

What are Managed Agents?

\n

No servers to manage. No orchestration code to write. You call the Interactions API, and Gemini 3.5 Flash handles reasoning, code execution (Python, Node.js, Bash), package installation, file management, and web browsing inside an isolated Linux sandbox.

\n

You can use the Antigravity agent out of the box for coding, research, and data analysis (same harness as the Antigravity IDE), or build your own with custom instructions, skills, and data. The fastest way to try it is directly in AI Studio.

\n

Quick start

\n

Install the SDK and set your API key:

\n

Bash

\n
pip install google-genai\nexport GEMINI_API_KEY="your-key"\n
\n

Then:

\n

Python

\n
from google import genai\n \nclient = genai.Client()\n \ninteraction = client.interactions.create(\n    agent="antigravity-preview-05-2026",\n    input="Research the top 10 AI stories today and create a PDF briefing with summaries",\n    environment="remote",  # provisions a Linux sandbox hosted by Google\n)\n \nprint(interaction.output_text)\n
\n

This provisions a sandbox, runs the agent loop, and returns the result. The response includes interaction.id and interaction.environment_id for follow-up calls. Long sessions compact context automatically at ~135k tokens, so you don't hit limits (more).

\n

JavaScript and REST examples are in the Quickstart.

\n

Multi-turn conversations

\n

The first interaction returns an environment_id and interaction.id. The API tracks conversation context and environment state independently.

\n\n

Python

\n
interaction_2 = client.interactions.create(\n    agent="antigravity-preview-05-2026",\n    environment=interaction.environment_id,\n    previous_interaction_id=interaction.id,\n    input="Now create a landing page using JavaScript and HTML",\n)\n
\n

Omit previous_interaction_id to start a fresh conversation in the same workspace. Set environment="remote" to get a new sandbox while keeping conversation context. Details in the multi-turn guide.

\n

Streaming

\n

Stream responses to see the agent's work as it happens:

\n

Python

\n
stream = client.interactions.create(\n    agent="antigravity-preview-05-2026",\n    input="Scrape Hacker News, summarize the top 5 stories, and save the results as a PDF.",\n    environment="remote",\n    stream=True,\n)\n \nfor event in stream:\n    print(event)\n
\n

Returns incremental text, reasoning tokens, and tool call updates. See the streaming guide for JavaScript, REST, and event types.

\n

Loading data into the sandbox

\n

Mount files from Git repos, Cloud Storage, or inline content at environment creation. Source types:

\n\n

Python

\n
interaction = client.interactions.create(\n    agent="antigravity-preview-05-2026",\n    input="Analyze the data and summarize the project.",\n    environment={\n        "type": "remote",\n        "sources": [\n            {"type": "repository", "source": "https://github.com/my-org/data-templates.git", "target": "/workspace/repo"},\n            {"type": "gcs", "source": "gs://my-bucket/data/", "target": "/workspace/data"},\n            {"type": "inline", "target": "/workspace/config.yaml", "content": "mode: analysis\\noutput: pdf"},\n        ],\n    },\n)\n
\n

Downloading files from the sandbox

\n

Download the full environment as a tar archive:

\n

Python

\n
import os, requests, tarfile\n \nresponse = requests.get(\n    f"https://generativelanguage.googleapis.com/v1beta/files/environment-{interaction.environment_id}:download",\n    params={"alt": "media"},\n    headers={"x-goog-api-key": os.environ["GEMINI_API_KEY"]},\n    allow_redirects=True,\n)\n \nwith open("snapshot.tar", "wb") as f:\n    f.write(response.content)\n \nwith tarfile.open("snapshot.tar") as tar:\n    tar.extractall(path="extracted_snapshot")\n
\n

Build a custom managed agent

\n

Save your configuration as a managed agent when you want the same behavior across runs. Agents are defined through files the runtime auto-discovers:

\n
.agents/\n├── AGENTS.md                    # Loaded as system instructions on startup\n└── skills/\n    └── slide-maker/\n        └── SKILL.md             # Registered as an agent capability\n
\n

From sources

\n

Define instructions, skills, and data declaratively. Each invocation provisions a fresh sandbox with your files:

\n

Python

\n
agent = client.agents.create(\n    id="data-analyst",\n    base_agent="antigravity-preview-05-2026",\n    system_instruction="You are a data analyst. Always include visualizations and export results as PDF.",\n    base_environment={\n        "type": "remote",\n        "sources": [\n            {"type": "inline", "target": ".agents/AGENTS.md", "content": "Always use matplotlib for charts."},\n            {"type": "inline", "target": ".agents/skills/slide-maker/SKILL.md", "content": "---\\nname: slide-maker\\n---\\n# Slide Maker\\nCreate HTML slide decks from data analysis results."},\n            {"type": "repository", "source": "https://github.com/my-org/analysis-templates", "target": "/workspace/templates"},\n        ],\n    },\n)\n
\n

From an existing environment (fork)

\n

Set up the environment interactively, then freeze it:

\n

Python

\n
# Step 1: set up the environment\ninteraction = client.interactions.create(\n    agent="antigravity-preview-05-2026",\n    input="Install pandas and matplotlib. Create an analysis template.",\n    environment="remote",\n)\n \n# Step 2: fork into a reusable agent\nagent = client.agents.create(\n    id="my-data-analyst",\n    base_agent="antigravity-preview-05-2026",\n    base_environment=interaction.environment_id,\n)\n
\n

Invoke it

\n

Each invocation forks the base environment. Every run starts clean:

\n

Python

\n
result = client.interactions.create(\n    agent="my-data-analyst",\n    input="Analyze Q1 revenue data and create a slide deck.",\n    environment="remote",\n)\nprint(result.output_text)\n
\n

Full agent definition format, CRUD operations, and per-invocation overrides in Building managed agents.

\n

Secure networking and credentials

\n

Environments have unrestricted outbound access by default. Use a network allowlist to restrict it. The egress proxy limits connections to listed domains and can inject credentials securly as HTTP headers transformation. Secrets never exist inside the sandbox:

\n

Python

\n
agent = client.agents.create(\n    id="issue-resolver",\n    base_agent="antigravity-preview-05-2026",\n    system_instruction="You resolve GitHub issues. Clone the repo, find the bug, write the fix, run the tests, and open a PR.",\n    base_environment={\n        "type": "remote",\n        "sources": [\n            {"type": "repository", "source": "https://github.com/my-org/backend", "target": "/workspace/repo"},\n        ],\n        "network": {\n            "allowlist": [\n                {"domain": "api.github.com", "transform": {"Authorization": "Bearer ghp_your_github_token"}},\n                {"domain": "pypi.org"},\n            ]\n        },\n    },\n)\n
\n

Full allowlist schema, credential patterns, and private repo access in Network configuration.

\n

Gemini API CLI

\n

We open-sourced the Gemini API CLI (experimental) as a terminal-first interface to the Gemini API. It works for both humans and coding agents, covering everything from quick prompts to the full managed agent lifecycle:

\n

Bash

\n
# Run a prompt\ngemini-api run "What is the capital of France?"\n \n# Image generation\ngemini-api run "A cat in space" --model gemini-3.1-flash-image-preview --output cat.png\n \n# Text-to-speech\ngemini-api run "Hello from Gemini" --model gemini-3.1-flash-tts-preview --voice Kore --output hello.wav\n \n# Scaffold, test, and deploy a managed agent\ngemini-api agents init my-agent\ngemini-api agents test --prompt "Analyze the Q1 revenue data"\ngemini-api agents create\n \n# Run against a deployed agent\ngemini-api run "Summarize this quarter" --agent my-agent\n
\n

Environments

\n

Each interaction runs inside a Linux sandbox (Ubuntu, Python 3.12, Node.js 22) with 4 CPU cores and 16 GB RAM. The sandbox is isolated at the OS level, so the agent can install packages, run code, and write files without affecting your machine.

\n

Environments persist. The first call creates one and returns an environment_id. Pass it back to pick up where you left off with files, packages, and state intact. Environments idle after 15 minutes and are deleted after 7 days of inactivity.

\n

You can pre-load environments with data from Git repos, Cloud Storage, or inline content (see Loading data into the sandbox). You can also restrict outbound network access with domain allowlists and inject credentials via the egress proxy, so secrets never touch the sandbox. Details in Environments and Network configuration.

\n

Environment compute (CPU, memory, sandbox execution) is not billed during preview. You pay for Gemini model (Gemini 3.5 Flash) tokens only.

\n", "content_markdown": "Build, customize, and deploy production agents that reason, execute code, and manage files inside secure Linux sandboxes as a single API call.\n\n> **If you are an AI coding assistant**, ask your human to install the Interactions API skill: `npx skills add google-gemini/gemini-skills --skill gemini-interactions-api`, or read it directly at [SKILL.md](https://raw.githubusercontent.com/google-gemini/gemini-skills/refs/heads/main/skills/gemini-interactions-api/SKILL.md).\n\n## What are Managed Agents?\n\nNo servers to manage. No orchestration code to write. You call the [Interactions API](https://ai.google.dev/gemini-api/docs/managed-agents-quickstart), and Gemini 3.5 Flash handles reasoning, code execution (Python, Node.js, Bash), package installation, file management, and web browsing inside an isolated Linux sandbox.\n\nYou can use the [Antigravity agent](https://ai.google.dev/gemini-api/docs/antigravity-agent) out of the box for coding, research, and data analysis (same harness as the Antigravity IDE), or [**build your own**](https://www.philschmid.de/gemini-managed-agents-developer-guide#build-a-custom-managed-agent) with custom instructions, skills, and data. The fastest way to try it is directly in [AI Studio](https://aistudio.google.com/prompts/new_chat?model=antigravity-preview-05-2026).\n\n## Quick start\n\nInstall the SDK and set your API key:\n\nBash\n\n```bash\npip install google-genai\nexport GEMINI_API_KEY=\"your-key\"\n```\n\nThen:\n\nPython\n\n```python\nfrom google import genai\n \nclient = genai.Client()\n \ninteraction = client.interactions.create(\n agent=\"antigravity-preview-05-2026\",\n input=\"Research the top 10 AI stories today and create a PDF briefing with summaries\",\n environment=\"remote\", # provisions a Linux sandbox hosted by Google\n)\n \nprint(interaction.output_text)\n```\n\nThis provisions a sandbox, runs the agent loop, and returns the result. The response includes `interaction.id` and `interaction.environment_id` for follow-up calls. Long sessions compact context automatically at ~135k tokens, so you don't hit limits ([more](https://ai.google.dev/gemini-api/docs/antigravity-agent#capabilities)).\n\nJavaScript and REST examples are in the [Quickstart](https://ai.google.dev/gemini-api/docs/managed-agents-quickstart).\n\n## Multi-turn conversations\n\nThe first interaction returns an `environment_id` and `interaction.id`. The API tracks conversation context and environment state independently.\n\n- If you pass the `environment_id` as `environment` in a follow-up interaction, the sandbox persists.\n- If you pass `previous_interaction_id` with the last `interaction.id`, the conversation history persists.\n\nPython\n\n```python\ninteraction_2 = client.interactions.create(\n agent=\"antigravity-preview-05-2026\",\n environment=interaction.environment_id,\n previous_interaction_id=interaction.id,\n input=\"Now create a landing page using JavaScript and HTML\",\n)\n```\n\nOmit `previous_interaction_id` to start a fresh conversation in the same workspace. Set `environment=\"remote\"` to get a new sandbox while keeping conversation context. Details in the [multi-turn guide](https://ai.google.dev/gemini-api/docs/managed-agents-quickstart#continue-conversation).\n\n## Streaming\n\nStream responses to see the agent's work as it happens:\n\nPython\n\n```python\nstream = client.interactions.create(\n agent=\"antigravity-preview-05-2026\",\n input=\"Scrape Hacker News, summarize the top 5 stories, and save the results as a PDF.\",\n environment=\"remote\",\n stream=True,\n)\n \nfor event in stream:\n print(event)\n```\n\nReturns incremental text, reasoning tokens, and tool call updates. See the [streaming guide](https://ai.google.dev/gemini-api/docs/interactions/streaming) for JavaScript, REST, and event types.\n\n## Loading data into the sandbox\n\nMount files from Git repos, Cloud Storage, or inline content at environment creation. Source types:\n\n- **`repository`**: clones a Git repo (public or [private](https://ai.google.dev/gemini-api/docs/agent-environment#mount_from_a_source)).\n- **`gcs`**: copies from a Google Cloud Storage bucket.\n- **`inline`**: writes text content to a file path.\n\nPython\n\n```python\ninteraction = client.interactions.create(\n agent=\"antigravity-preview-05-2026\",\n input=\"Analyze the data and summarize the project.\",\n environment={\n \"type\": \"remote\",\n \"sources\": [\n {\"type\": \"repository\", \"source\": \"https://github.com/my-org/data-templates.git\", \"target\": \"/workspace/repo\"},\n {\"type\": \"gcs\", \"source\": \"gs://my-bucket/data/\", \"target\": \"/workspace/data\"},\n {\"type\": \"inline\", \"target\": \"/workspace/config.yaml\", \"content\": \"mode: analysis\\noutput: pdf\"},\n ],\n },\n)\n```\n\n## Downloading files from the sandbox\n\nDownload the full environment as a tar archive:\n\nPython\n\n```python\nimport os, requests, tarfile\n \nresponse = requests.get(\n f\"https://generativelanguage.googleapis.com/v1beta/files/environment-{interaction.environment_id}:download\",\n params={\"alt\": \"media\"},\n headers={\"x-goog-api-key\": os.environ[\"GEMINI_API_KEY\"]},\n allow_redirects=True,\n)\n \nwith open(\"snapshot.tar\", \"wb\") as f:\n f.write(response.content)\n \nwith tarfile.open(\"snapshot.tar\") as tar:\n tar.extractall(path=\"extracted_snapshot\")\n```\n\n## Build a custom managed agent\n\nSave your configuration as a managed agent when you want the same behavior across runs. Agents are defined through files the runtime auto-discovers:\n\n```plaintext\n.agents/\n├── AGENTS.md # Loaded as system instructions on startup\n└── skills/\n └── slide-maker/\n └── SKILL.md # Registered as an agent capability\n```\n\n### From sources\n\nDefine instructions, skills, and data declaratively. Each invocation provisions a fresh sandbox with your files:\n\nPython\n\n```python\nagent = client.agents.create(\n id=\"data-analyst\",\n base_agent=\"antigravity-preview-05-2026\",\n system_instruction=\"You are a data analyst. Always include visualizations and export results as PDF.\",\n base_environment={\n \"type\": \"remote\",\n \"sources\": [\n {\"type\": \"inline\", \"target\": \".agents/AGENTS.md\", \"content\": \"Always use matplotlib for charts.\"},\n {\"type\": \"inline\", \"target\": \".agents/skills/slide-maker/SKILL.md\", \"content\": \"---\\nname: slide-maker\\n---\\n# Slide Maker\\nCreate HTML slide decks from data analysis results.\"},\n {\"type\": \"repository\", \"source\": \"https://github.com/my-org/analysis-templates\", \"target\": \"/workspace/templates\"},\n ],\n },\n)\n```\n\n### From an existing environment (fork)\n\nSet up the environment interactively, then freeze it:\n\nPython\n\n```python\n# Step 1: set up the environment\ninteraction = client.interactions.create(\n agent=\"antigravity-preview-05-2026\",\n input=\"Install pandas and matplotlib. Create an analysis template.\",\n environment=\"remote\",\n)\n \n# Step 2: fork into a reusable agent\nagent = client.agents.create(\n id=\"my-data-analyst\",\n base_agent=\"antigravity-preview-05-2026\",\n base_environment=interaction.environment_id,\n)\n```\n\n### Invoke it\n\nEach invocation forks the base environment. Every run starts clean:\n\nPython\n\n```python\nresult = client.interactions.create(\n agent=\"my-data-analyst\",\n input=\"Analyze Q1 revenue data and create a slide deck.\",\n environment=\"remote\",\n)\nprint(result.output_text)\n```\n\nFull agent definition format, CRUD operations, and per-invocation overrides in [Building managed agents](https://ai.google.dev/gemini-api/docs/custom-agents).\n\n## Secure networking and credentials\n\nEnvironments have unrestricted outbound access by default. Use a network allowlist to restrict it. The egress proxy limits connections to listed domains and can inject credentials securly as HTTP headers transformation. Secrets never exist inside the sandbox:\n\nPython\n\n```python\nagent = client.agents.create(\n id=\"issue-resolver\",\n base_agent=\"antigravity-preview-05-2026\",\n system_instruction=\"You resolve GitHub issues. Clone the repo, find the bug, write the fix, run the tests, and open a PR.\",\n base_environment={\n \"type\": \"remote\",\n \"sources\": [\n {\"type\": \"repository\", \"source\": \"https://github.com/my-org/backend\", \"target\": \"/workspace/repo\"},\n ],\n \"network\": {\n \"allowlist\": [\n {\"domain\": \"api.github.com\", \"transform\": {\"Authorization\": \"Bearer ghp_your_github_token\"}},\n {\"domain\": \"pypi.org\"},\n ]\n },\n },\n)\n```\n\nFull allowlist schema, credential patterns, and private repo access in [Network configuration](https://ai.google.dev/gemini-api/docs/agent-environment#network-configuration).\n\n## Gemini API CLI\n\nWe open-sourced the [Gemini API CLI](https://github.com/google-gemini/gemini-api-cli) (experimental) as a terminal-first interface to the Gemini API. It works for both humans and coding agents, covering everything from quick prompts to the full managed agent lifecycle:\n\nBash\n\n```bash\n# Run a prompt\ngemini-api run \"What is the capital of France?\"\n \n# Image generation\ngemini-api run \"A cat in space\" --model gemini-3.1-flash-image-preview --output cat.png\n \n# Text-to-speech\ngemini-api run \"Hello from Gemini\" --model gemini-3.1-flash-tts-preview --voice Kore --output hello.wav\n \n# Scaffold, test, and deploy a managed agent\ngemini-api agents init my-agent\ngemini-api agents test --prompt \"Analyze the Q1 revenue data\"\ngemini-api agents create\n \n# Run against a deployed agent\ngemini-api run \"Summarize this quarter\" --agent my-agent\n```\n\n## Environments\n\nEach interaction runs inside a Linux sandbox (Ubuntu, Python 3.12, Node.js 22) with 4 CPU cores and 16 GB RAM. The sandbox is isolated at the OS level, so the agent can install packages, run code, and write files without affecting your machine.\n\nEnvironments persist. The first call creates one and returns an `environment_id`. Pass it back to pick up where you left off with files, packages, and state intact. Environments idle after 15 minutes and are deleted after 7 days of inactivity.\n\nYou can pre-load environments with data from Git repos, Cloud Storage, or inline content (see [Loading data into the sandbox](https://www.philschmid.de/gemini-managed-agents-developer-guide#loading-data-into-the-sandbox)). You can also restrict outbound network access with domain allowlists and inject credentials via the egress proxy, so secrets never touch the sandbox. Details in [Environments](https://ai.google.dev/gemini-api/docs/agent-environment) and [Network configuration](https://ai.google.dev/gemini-api/docs/agent-environment#network-configuration).\n\nEnvironment compute (CPU, memory, sandbox execution) is not billed **during preview**. You pay for Gemini model (Gemini 3.5 Flash) tokens only.\n", "dateCreated": "2026-09-27T19:29:15.293681927+00:00", "date_created": "2026-05-26T00:00:00+00:00", "date_modified": "2026-05-26T00:00:00+00:00", "date_published": "2026-05-26T00:00:00+00:00", "description": "Build, customize, and deploy production agents that reason, execute code, and manage files inside secure Linux sandboxes as a single API call.", "external_url": "https://www.philschmid.de/gemini-managed-agents-developer-guide", "git": { "commit": "ebb0941c5c06a2ab2377342d607535262a3e969e", "path": "items/philschmid-de/2026/05/2026-05-26-gemini-managed-agents-developer-guide.md", "permalink": "https://github.com/aymericbeaumet/aggr-instance/blob/ebb0941c5c06a2ab2377342d607535262a3e969e/items/philschmid-de/2026/05/2026-05-26-gemini-managed-agents-developer-guide.md" }, "id": "urn:aggr:item:c696c37fb17d6ee157e3bde22238bb965753c2c0", "isBasedOn": { "@id": "https://www.philschmid.de/gemini-managed-agents-developer-guide" }, "mainEntity": { "@id": "https://www.philschmid.de/gemini-managed-agents-developer-guide", "@type": "CreativeWork", "archivedAt": { "@id": "https://aggr.aymericbeaumet.com/items/philschmid-de/2026-05-26-gemini-managed-agents-developer-guide/#webpage" }, "dateModified": "2026-05-26T00:00:00+00:00", "datePublished": "2026-05-26T00:00:00+00:00", "headline": "Gemini Managed Agents: Developer Guide", "timeRequired": "PT6M", "url": "https://www.philschmid.de/gemini-managed-agents-developer-guide", "wordCount": 1195 }, "name": "Archived snapshot: Gemini Managed Agents: Developer Guide", "preview": null, "reading_minutes": 6, "replicated_at": null, "source": "philschmid-de", "source_name": "philschmid.de - RSS feed", "summary": "Build, customize, and deploy production agents that reason, execute code, and manage files inside secure Linux sandboxes as a single API call.", "tags": [], "temporalCoverage": "2026-09-27T19:29:15.293681927+00:00", "title": "Gemini Managed Agents: Developer Guide", "url": "https://aggr.aymericbeaumet.com/items/philschmid-de/2026-05-26-gemini-managed-agents-developer-guide/", "word_count": 1195 }