Add thinking mode for ai logic hybrid by VinayGuthal · Pull Request #8678 · firebase/firebase-android-sdk

VinayGuthal · 2026-10-07T15:55:14Z

Adds support for thinking mode (enableThinking) when generating content using on-device and hybrid inference modes.

Summary of Changes

firebase-ai-ondevice-interop:
- Added enableThinking: Boolean? to GenerateContentRequest.
- Added thoughtProcess: List<Candidate> to GenerateContentResponse.
- Added isThinkingModeAvailable(): Boolean to GenerativeModel.
- Updated api.txt.
firebase-ai-ondevice:
- Implemented GenerativeModelImpl.isThinkingModeAvailable() delegating to ML Kit's GenerativeModel.isThinkingModeAvailable().
- Updated Converters.kt to pass enableThinking to the ML Kit request and map thoughtProcess candidates to the interop GenerateContentResponse.
firebase-ai:
- Added enableThinking: Boolean? = null to OnDeviceConfig and updated api.txt.
- Updated OnDeviceGenerativeModelProvider to check onDeviceModel.isThinkingModeAvailable() when enableThinking == true (throwing FirebaseAIOnDeviceNotAvailableException so PREFER_ON_DEVICE falls back to cloud when unsupported) and forward enableThinking to the on-device request.
- Updated Candidate and OnDeviceGenerativeModelProvider response mapping to convert interop thoughtProcess candidates into TextPart entries with isThought = true (populating GenerateContentResponse.thoughtSummary for both unary and streaming responses).
Tests & Changelogs: Added unit tests in ConvertersTest, GenerativeModelImplTest, and OnDeviceGenerativeModelProviderTests, and updated CHANGELOG.md across all three modules.

Tested at firebase/quickstart-android#2848

gemini-code-assist · 2026-10-07T15:55:19Z

Using Gemini Code Assist

The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.

Invoking Gemini

You can request assistance from Gemini at any point by creating a comment using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands on the current page.

Feature	Command	Description
Code Review	`/gemini review`	Performs a code review for the current pull request in its current state.
Pull Request Summary	`/gemini summary`	Provides a summary of the current pull request in its current state.
Comment	@gemini-code-assist	Responds in comments when explicitly tagged, both in pull request comments and review comments.
Help	`/gemini help`	Displays a list of available commands.

Customization

To customize the Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.

Limitations & Feedback

Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counterproductive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for GitHub and other Google products, sign up here.

github-actions · 2026-10-07T15:55:34Z

📝 PRs merging into main branch

Our main branch should always be in a releasable state. If you are working on a larger change, or if you don't want this change to see the light of the day just yet, consider using a feature branch first, and only merge into the main branch when the code complete and ready to be released.

VinayGuthal · 2026-10-08T14:35:05Z

/gemini review

gemini-code-assist

Code Review

This pull request adds support for thinking mode in on-device content generation, introducing configuration options like enableThinking, tracking the model's reasoning process via thoughtProcess, and adding availability checks. The review feedback highlights two key improvements: first, updating the candidate mapping logic to accept a list of thoughts rather than a single thought to prevent reasoning steps from being discarded when there is only one candidate; second, disabling the thinking mode check during token counting to avoid unnecessary failures, as tokenization does not execute the model's thinking process.

…nerativemodel/OnDeviceGenerativeModelProvider.kt Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>

VinayGuthal · 2026-10-08T15:30:38Z

/gemini review

gemini-code-assist

Code Review

This pull request introduces support for thinking mode in on-device content generation, adding enableThinking to configuration and request classes, thoughtProcess to response classes, and isThinkingModeAvailable to the generative model interface. Feedback on these changes highlights a critical issue in OnDeviceGenerativeModelProvider.toPublicResponse where sequential thought steps are incorrectly sliced across multiple candidates or mapped to separate candidates when the response candidates list is empty. Refactoring is suggested to associate the entire thoughtProcess with each candidate, which would also allow the removal of the now-redundant fromInteropThought helper function.

VinayGuthal · 2026-10-08T15:50:06Z

/gemini review

gemini-code-assist

Code Review

This pull request introduces support for "thinking mode" in on-device content generation by adding enableThinking to request configurations, thoughtProcess to responses, and isThinkingModeAvailable to model interfaces across the firebase-ai and firebase-ai-ondevice modules. Feedback recommends wrapping the initialization block of generateContentStream in a try-catch to ensure setup exceptions are properly wrapped in FirebaseAIException, and filtering out empty thoughts in fromInteropThought to maintain consistency with fromInterop.

…pe/Candidate.kt Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>

VinayGuthal · 2026-10-08T16:14:54Z

/gemini review

gemini-code-assist

Code Review

This pull request introduces support for thinking mode in on-device content generation. It adds enableThinking to request configurations, exposes thoughtProcess in responses, and provides a method to check if thinking mode is available on the model. These changes are propagated through the interop layers, public APIs, and converters, accompanied by corresponding unit tests. Feedback on the changes suggests removing a fragile assertion in ConvertersTest.kt that checks the default value of enableThinking when not explicitly set, as it contradicts the test's stated design principle.

rlazo · 2026-10-09T13:46:36Z

@@ -1,5 +1,6 @@
 # Unreleased

+- [feature] Added `enableThinking` to `GenerateContentRequest`, `thoughtProcess` to `GenerateContentResponse`, and `isThinkingModeAvailable` to `GenerativeModel` to support thinking mode.


Max 80 chars per line. Also, add PR number

rlazo · 2026-10-09T13:48:35Z

      topK = this@toMlKit.topK
      seed = this@toMlKit.seed
      systemInstruction = this@toMlKit.systemInstruction?.toMlKitSystemInstruction()
+      this@toMlKit.enableThinking?.let { enableThinking = it }


This statement logic is reversed to all the others (not a direct assignment)

rlazo · 2026-10-09T13:55:00Z

@@ -1,5 +1,6 @@
 # Unreleased

+- [feature] Added support for thinking mode in on-device content generation requests and responses.


Add PR number

rlazo · 2026-10-10T00:13:51Z

+            nonEmptyThoughts.forEach { thought ->
+              part(TextPart(text = thought.text, isThought = true, thoughtSignature = null))
+            }
+            if (candidate.text.isNotEmpty() || nonEmptyThoughts.isEmpty()) {


could you walk me through this part? All thoughts are included in the content, but if there are no thoughts, an empty text is used?

rlazo · 2026-10-10T00:15:54Z

@@ -1,5 +1,6 @@
 # Unreleased

+- [feature] Added `enableThinking` to `OnDeviceConfig` to support [thinking mode](https://firebase.google.com/docs/ai-logic/thinking) in on-device and hybrid inference modes, with automatic cloud fallback when the on-device model does not support thinking.


Max length 80, include PR number

add thinking mode

7718978

gemini-code-assist Bot reviewed Oct 8, 2026

View reviewed changes

VinayGuthal and others added 3 commits October 8, 2026 10:37

Merge branch 'main' into add_thinking_mode

3011c8f

Update ai-logic/firebase-ai/src/main/kotlin/com/google/firebase/ai/ge…

ea45176

…nerativemodel/OnDeviceGenerativeModelProvider.kt Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>

update

7da28f8

VinayGuthal changed the title ~~Add thinking mode.~~ Add thinking mode for ai logic hybrid Oct 8, 2026

gemini-code-assist Bot reviewed Oct 8, 2026

View reviewed changes

Comment thread ...ai/src/main/kotlin/com/google/firebase/ai/generativemodel/OnDeviceGenerativeModelProvider.kt

Comment thread ai-logic/firebase-ai/src/main/kotlin/com/google/firebase/ai/type/Candidate.kt Outdated

update

d9a60b2

gemini-code-assist Bot reviewed Oct 8, 2026

View reviewed changes

Comment thread ...ai/src/main/kotlin/com/google/firebase/ai/generativemodel/OnDeviceGenerativeModelProvider.kt

Comment thread ai-logic/firebase-ai/src/main/kotlin/com/google/firebase/ai/type/Candidate.kt

VinayGuthal and others added 2 commits October 8, 2026 12:04

Update ai-logic/firebase-ai/src/main/kotlin/com/google/firebase/ai/ty…

42fb811

…pe/Candidate.kt Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>

apply suggestion

8ca2330

gemini-code-assist Bot reviewed Oct 8, 2026

View reviewed changes

Comment thread ai-logic/firebase-ai-ondevice/src/test/kotlin/com/google/firebase/ai/ondevice/ConvertersTest.kt

VinayGuthal requested a review from rlazo October 9, 2026 13:37

rlazo reviewed Oct 10, 2026

View reviewed changes

		@@ -1,5 +1,6 @@
		# Unreleased

		- [feature] Added `enableThinking` to `GenerateContentRequest`, `thoughtProcess` to `GenerateContentResponse`, and `isThinkingModeAvailable` to `GenerativeModel` to support thinking mode.

		@@ -1,5 +1,6 @@
		# Unreleased

		- [feature] Added support for thinking mode in on-device content generation requests and responses.

		@@ -1,5 +1,6 @@
		# Unreleased

		- [feature] Added `enableThinking` to `OnDeviceConfig` to support [thinking mode](https://firebase.google.com/docs/ai-logic/thinking) in on-device and hybrid inference modes, with automatic cloud fallback when the on-device model does not support thinking.

Conversation

VinayGuthal commented Oct 7, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary of Changes

Uh oh!

gemini-code-assist Bot commented Oct 7, 2026

Uh oh!

github-actions Bot commented Oct 7, 2026

📝 PRs merging into main branch

Uh oh!

VinayGuthal commented Oct 8, 2026

Uh oh!

gemini-code-assist Bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

Uh oh!

Uh oh!

Uh oh!

VinayGuthal commented Oct 8, 2026

Uh oh!

gemini-code-assist Bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

Uh oh!

Uh oh!

VinayGuthal commented Oct 8, 2026

Uh oh!

gemini-code-assist Bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

Uh oh!

Uh oh!

VinayGuthal commented Oct 8, 2026

Uh oh!

gemini-code-assist Bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

Uh oh!

rlazo Oct 9, 2026

Choose a reason for hiding this comment

Uh oh!

rlazo Oct 9, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

rlazo Oct 9, 2026

Choose a reason for hiding this comment

Uh oh!

rlazo Oct 10, 2026

Choose a reason for hiding this comment

Uh oh!

rlazo Oct 10, 2026

Choose a reason for hiding this comment

Uh oh!

This branch has not been deployed

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

VinayGuthal commented Oct 7, 2026 •

edited

Loading