Will AI Logic context caching be supported in the iOS SDK?

[revivn-brett]

Hi! I see that the latest Firebase release 12.9.0 has introduced response metadata for context caching when using Firebase AI Logic. Is this a sign that full context caching support will be rolled into the iOS SDK soon? This is important to us in deciding whether to choose the native SDK for making AI requests versus a more convoluted backend solution, as we have found context caching pretty important in keeping costs down. Thanks for any insight you can provide.

[paulb777]

Are you asking about explicit context caching support in addition to implicit context caching? https://ai.google.dev/gemini-api/docs/caching What SDK APIs would you like to have?

[revivn-brett]

Yes apologies, I was referring to explicit context caching as in the docs you linked. As in those docs, at minimum it would be great to be able to create an explict cache and then reference that cache in GenerationConfig (or equivalent) passed in via the generateContent method.

[marb2000]

Hi there!

We have added the ability to use an existing explicit caching in Firebase AI Logic. You should create the explicit caching using the Gemini API directly or via the Google Gen AI SDKs. For example, this is way I created a explicit caching in a Colab.

# @title Create Explicit Content Cache using `google-genai` SDK

# 1. Install the Google Gen AI Python SDK
# !pip install --upgrade --quiet google-genai

import sys
from google.colab import auth
from google import genai
from google.genai.types import CreateCachedContentConfig, Part

# 2. Authenticate your environment
print("Authenticating user...")
auth.authenticate_user()

# 3. Project Configuration
PROJECT_ID = "ai-logic-demos" # @param {type:"string"}
LOCATION = "us-central1"      # @param {type:"string"}
MODEL_ID = "gemini-2.5-flash" # @param {type:"string"}

# Initialize the client (Vertex AI backend)
print(f"Initializing GenAI Client for project '{PROJECT_ID}'...")
client = genai.Client(vertexai=True, project=PROJECT_ID, location=LOCATION)

# 4. Define Content and TTL
# Replace with your GCS URI or local file content
GCS_FILE_URI = "gs://ai-logic-demos.firebasestorage.app/Detailed_Human_Personas.md"
gcs_file_part = Part.from_uri(file_uri=GCS_FILE_URI, mime_type="text/plain")

SYSTEM_PROMPT = "Just respond to questions based on the content I provided you."
CACHE_DISPLAY_NAME = "characters-content"
TTL_SECONDS = "3600s" # 1 Hour

# 5. Create the Cache
print("Creating explicit content cache...")
gcs_cache = client.caches.create(
    model=MODEL_ID,
    config=CreateCachedContentConfig(
        system_instruction=SYSTEM_PROMPT,
        contents=[gcs_file_part],
        display_name=CACHE_DISPLAY_NAME,
        ttl=TTL_SECONDS
    ),
)

# 6. Output the Resource Name
# Save this ID! You will use it in your Firebase Prompt Template.
print(f"Cache Created successfully.")
print(f"Cache Name: {gcs_cache.name}")
# Example Output: "projects/861083271982/locations/us-central1/cachedContents/4545031458888089600"

Once the cache is created, you can reference it within your server prompt template using the cachedContent property

{{cachedContent id="YOUR_CACHE_RESOURCE_NAME"}}

{{role "user"}}
{{userPrompt}}

Example:

{{cachedContent id="projects/861083271982/locations/us-central1/cachedContents/4545031458888089600"}}

{{role "user"}}
{{userPrompt}}

If the specified cache ID is invalid or has expired (past its TTL), the request will fail with an "Explicit content cache doesn't exist" error.
The id parameter can also be a dynamic variable. For example, {{cachedContent id=someVariable}} would allow specifying the cache ID as an input to TemplateGenerateContent.

There is not support for explicit caching via direct prompts, just using server prompt templates.

[revivn-brett]

Hi Miguel, thanks for the reply. I was not aware of prompt templates! This appears to be a Preview feature so it's not something I could use for production yet; do y'all have any idea when it might leave Preview status?