GCS Media Storage

Offload agent media to Google Cloud Storage and persist a MediaReference after successful storage.

GCSMediaStorage writes media to Google Cloud Storage and replaces inline bytes with a MediaReference after a successful upload.

Media offload is fail-open. If the GCS upload fails, the run can still succeed and persist the inline media instead of a MediaReference.

media_storage_gcs.py

"""
GCS Media Storage
=================

Demonstrates GCSMediaStorage, which offloads media to Google Cloud Storage and keeps only a
MediaReference in the database. URL-only media is skipped unless persist_remote_urls=True.

Signing a URL needs a service-account private key, so under application-default credentials
nothing is signed and AgentOS streams the bytes through the /media route instead.

Requirements:
- uv pip install 'agno[gcs]'
- `gcloud auth application-default login`, or a service-account JSON via credentials_path
- Set MEDIA_GCS_BUCKET to a bucket you own; GCP_PROJECT to set the project
"""

import os

import httpx
from agno.agent import Agent
from agno.db.sqlite import SqliteDb
from agno.media import Image
from agno.media.storage import GCSMediaStorage
from agno.models.openai import OpenAIResponses

# ---------------------------------------------------------------------------
# Setup
# ---------------------------------------------------------------------------
IMAGE_URL = "https://thumbs.dreamstime.com/b/mountain-landscape-pieniny-national-park-foot-tatra-mountains-mountain-landscape-pieniny-national-park-437239881.jpg?w=768"

# A bucket you do not own makes every upload fail, and offload falls back to inline
# base64 — the run still succeeds, so the failure is easy to miss. Ask for the bucket
# up front instead.
bucket = os.getenv("MEDIA_GCS_BUCKET")
if not bucket:
    raise ValueError("MEDIA_GCS_BUCKET must be set to a GCS bucket you own")

# ---------------------------------------------------------------------------
# Approach 1: Pre-download the media yourself and send bytes.
# URL-only media is skipped by default.
# ---------------------------------------------------------------------------

storage = GCSMediaStorage(
    bucket=bucket,
    project=os.getenv("GCP_PROJECT"),
    prefix="agno/media/",
    presigned_url_expiry=3600,  # 1 hour
)

# ---------------------------------------------------------------------------
# Create the Agent
# ---------------------------------------------------------------------------
agent = Agent(
    model=OpenAIResponses(id="gpt-5.5"),
    media_storage=storage,
    db=SqliteDb(db_file="tmp/data.db"),
)

# ---------------------------------------------------------------------------
# Approach 2: Use the flag persist_remote_urls=True.
# This will download every URL-only media automatically and store it in GCS.
# ---------------------------------------------------------------------------

storage_with_persist = GCSMediaStorage(
    bucket=bucket,
    project=os.getenv("GCP_PROJECT"),
    prefix="agno/media/",
    presigned_url_expiry=3600,  # 1 hour
    persist_remote_urls=True,
)

agent_with_persist = Agent(
    model=OpenAIResponses(id="gpt-5.5"),
    media_storage=storage_with_persist,
    db=SqliteDb(db_file="tmp/data.db"),
)

# ---------------------------------------------------------------------------
# Run the Agents
# ---------------------------------------------------------------------------
if __name__ == "__main__":
    # Download image content first so media storage can offload it to GCS.
    # mime_type gives the stored object its file extension and Content-Type.
    image_bytes = httpx.get(IMAGE_URL, follow_redirects=True).content

    agent.print_response(
        "What do you see in this image?",
        images=[Image(content=image_bytes, format="jpeg", mime_type="image/jpeg")],
    )

    # URL-only media is NOT stored in GCS by default — it is skipped during offload.
    agent.print_response(
        "What do you see in this image?",
        images=[Image(url=IMAGE_URL)],
    )

    # URL-only images are automatically downloaded and stored when persist_remote_urls=True
    agent_with_persist.print_response(
        "What do you see in this image?",
        images=[Image(url=IMAGE_URL)],
    )

Run the Example

1. Set up your virtual environment

   Mac

   uv venv --python 3.12
   source .venv/bin/activate
   
2. Install dependencies

   uv pip install -U "agno[gcs]" openai sqlalchemy
   
3. Export environment variables

   Mac/Linux

   export GCP_PROJECT="your_gcp_project_here"
   export MEDIA_GCS_BUCKET="your_media_gcs_bucket_here"
   export OPENAI_API_KEY="your_openai_api_key_here"
   
4. Authenticate with Google Cloud

   Sign in with Application Default Credentials:

   gcloud auth application-default login
   
5. Run the example

   Save the code above as media_storage_gcs.py, then run:

   python media_storage_gcs.py
   

Full source: cookbook/06_storage/08_media_storage_gcs.py