.Net: Add CachedContent Property to GeminiPromptExecutionSettings

[davidpene]

Motivation and Context

Why is this change required?
This change introduces a new CachedContent field to the GeminiPromptExecutionSettings class. The addition enables context caching, a feature that optimizes the handling of repeated static content in requests with high input token counts.

What problem does it solve?
Repeatedly processing substantial, static context data (e.g., lengthy documents, audio files, or video files) in requests can be resource-intensive and costly. Context caching addresses this by allowing the reuse of shared context across multiple requests, improving performance and cost-efficiency.

What scenario does it contribute to?
Context caching is particularly well-suited for scenarios where a substantial initial context is repeatedly referenced by shorter requests. Use cases include:

• Chatbots with extensive system instructions: Allows reuse of complex system configurations across multiple user interactions.
• Repetitive analysis of lengthy video files: Enables efficient analysis by caching video-related metadata or transcriptions.
• Recurring queries against large document sets: Improves efficiency for workflows requiring repeated access to large document collections.
• Frequent code repository analysis or bug fixing: Facilitates the reuse of large codebases as cached context for debugging or analysis tasks.

No open issues are linked to this change.

---

Description

This PR adds a CachedContent field to the GeminiPromptExecutionSettings class. The field allows users to reference cached context items such as text blocks, audio files, or video files in prompt requests.

Key features:

• The minimum size of a context cache is 32,768 tokens.
• Cached content is retained for a default duration of 60 minutes, with the option to configure expiration.
• Cached content is billed efficiently: the initial creation call is charged at the standard rate, while subsequent references to the cache are billed at a reduced rate.

By enabling context caching, users can optimize cost and resource usage for workflows requiring substantial and repeat context data.

---

Contribution Checklist

• The code builds clean without any errors or warnings
• The PR follows the SK Contribution Guidelines and the pre-submission formatting script raises no violations
• All unit tests pass, and I have added new tests where possible
• I didn't break anyone 😄

---

Demo

---

Notes

1. The CachedContent is only available via v1beta1 endpoint
2. Overview of context caching in Gemini
3. cachedContent field in the body for the v1beta1 endpoint

[davidpene]

@microsoft-github-policy-service agree company="Carepatron"

[davidpene]

@RogerBarreto / @markwallace-microsoft / @dmytrostruk would love to hear your thoughts.

[RogerBarreto]

@davidpene Thanks very much for the contribution, suggest adding missing UnitTests ensuring the new request property is behaving correctly when serializing, deserializing and going over the http, suggest adding an integration test ensuring it as well (keep the integration test Skipped for the project but ensure it works as expected when runningh locally)

For reference check:

https://github.com/microsoft/semantic-kernel/blob/main/dotnet/src/Connectors/Connectors.Google.UnitTests/Core/Gemini/GeminiRequestTests.cs

[davidpene]

Appreciate the feedback @RogerBarreto - all done ✅
Let me know if there's anything else.