feat: @langchain/google by hntrl · Pull Request #10000 · langchain-ai/langchainjs

Hunter Lovell (hntrl) · 2026-02-11T08:08:24Z

Summary

Add new @langchain/google package -- a unified Google/Gemini integration that will replace @langchain/google-genai, @langchain/google-common, and the Vertex AI family of packages.

Huge huge huge shoutout to Allen Firstenberg (@afirstenberg) for driving a lot of this feature work

Why a new package?

The existing Google integration stack has accumulated significant technical debt:

@langchain/google-genai depends on a long-deprecated SDK from Google (@google/generative-ai). That SDK has been superseded by @google/genai, but migrating to it would be a breaking change with limited upside -- we'd still be coupling to Google's SDK release cycle and its opinionated abstractions.
@langchain/google-common / Vertex AI packages had to adopt an awkward multi-package split (google-common, google-gauth, google-webauth, google-vertexai, google-vertexai-web) to accommodate web and edge environments. This fragmentation makes it confusing for users to know which package to install, duplicates logic across packages, and creates a maintenance burden.
Hand-coded types in google-common drifted from the actual Gemini API surface over time, leading to broken fields, missing properties, and constant churn to keep them in sync with Google's rapid iteration.

Usage

The package exposes two entrypoints. The constructor takes the model name as the first positional argument, with an optional params object:

@langchain/google -- for web, edge, and serverless environments (Cloudflare Workers, Vercel Edge, Deno, browsers). Uses the Web Fetch API and authenticates via API key only.

import { ChatGoogle } from "@langchain/google";

const model = new ChatGoogle("gemini-2.5-flash", {
  apiKey: "...", // or set GOOGLE_API_KEY env var
});

@langchain/google/node -- for Node.js environments. Extends the base entrypoint with google-auth-library support, enabling Vertex AI via Application Default Credentials (ADC), service account JSON, and other Node-native auth flows.

import { ChatGoogle } from "@langchain/google/node";

// Google AI with API key (reads GOOGLE_API_KEY env var if not provided)
const aiStudio = new ChatGoogle("gemini-2.5-flash");

// Vertex AI with ADC (auto-detected)
const vertex = new ChatGoogle("gemini-2.5-flash", {
  vertexai: true,
  project: "my-gcp-project",
  location: "us-central1",
});

Both entrypoints export the same ChatGoogle class name, so switching between them is just an import path change.

Versioning and stability

This package is being published at 0.1.0 to signal that it is an early release. While we have confidence in the core functionality (chat, streaming, tool calling, reasoning, multimodal), the API surface may evolve as we gather feedback and close the remaining feature gaps relative to the existing packages. We intend to promote to 1.0 once we've validated the API through user feedback and reached feature parity with the existing packages.

Support policy for existing packages

The existing @langchain/google-genai, @langchain/google-vertexai, and related packages will continue to receive maintenance updates (bug fixes, security patches) for the foreseeable future. We are not deprecating them with this release. Once @langchain/google reaches feature parity, we will publish a migration guide and announce a deprecation timeline with adequate notice.

Changes

New package: `libs/providers/langchain-google/`

Area	Files	Description
Chat model	`base.ts`, `node.ts`, `index.ts`, `types.ts`	`ChatGoogle` extending `BaseChatModel` with tool calling, streaming, structured output, and multimodal support
API types	`api-types.ts`, `typegen.jsonc`	Auto-generated `Gemini` namespace from the OpenAPI spec, plus module augmentations for utility types and spec gaps
Converters	`converters/messages.ts`, `converters/params.ts`, `converters/tools.ts`	Bidirectional conversion between LangChain messages/tools and Gemini wire format
Clients	`clients/index.ts`, `clients/node.ts`	HTTP client with platform-specific auth (Google AI API key vs Vertex AI OAuth/ADC)
Utilities	`utils/errors.ts`, `utils/gcp-auth.ts`, `utils/handler.ts`, `utils/misc.ts`, `utils/stream.ts`	Error handling, GCP auth helpers, SSE stream parsing
Tests	`tests/index.test.ts`, `tests/index.int.test.ts`	40 unit tests (reasoning config across model families), integration tests

`@langchain/core` changes

Added ChatGoogleTranslator block translator (block_translators/google.ts) and registered it under the "google" provider key.

changeset-bot · 2026-02-11T08:08:31Z

🦋 Changeset detected

Latest commit: 77f1689

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 14 packages

Name	Type
@langchain/google	Minor
@langchain/core	Patch
@langchain/community	Patch
langchain	Patch
@langchain/anthropic	Patch
@langchain/google-cloud-sql-pg	Patch
@langchain/google-common	Patch
@langchain/google-genai	Patch
@langchain/google-webauth	Patch
@langchain/model-profiles	Patch
@langchain/standard-tests	Patch
@langchain/google-gauth	Patch
@langchain/google-vertexai-web	Patch
@langchain/google-vertexai	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

Christian Bromann (christian-bromann)

Thanks for the contribution! This is an excellent and comprehensive implementation of a unified Google/Gemini integration. The code is well-structured with proper abstractions, thorough error handling, and extensive documentation. The approach of consolidating the fragmented package ecosystem is very valuable. A few edge case suggestions and improvements below.

Christian Bromann (christian-bromann) · 2026-02-11T23:52:11Z

libs/providers/langchain-google/src/clients/index.ts

+    }
+    return fetch(request);
+  }
+}


The authentication priority logic here could cause issues. When both apiKey and credentials are provided, the function only checks this.params.apiKey but should also check the resolved this.apiKey field:

Suggested change

}

fetch(request: Request): Promise<Response> {

if (this.apiKey) {

request.headers.set(GOOGLE_API_KEY_HEADER, this.apiKey);

} else if (this.credentials) {

request.headers.set(

GCP_API_KEY_HEADER,

`Bearer ${getGCPCredentialsAccessToken(this.credentials)}`

);

}

return fetch(request);

}

This ensures consistent behavior with the hasApiKey() method.

libs/providers/langchain-google/src/converters/tools.ts

Christian Bromann (christian-bromann) · 2026-02-11T23:52:12Z

libs/providers/langchain-google/src/utils/errors.ts

+    isInstance(obj: unknown): obj is this {
+      return (
+        typeof obj === "object" &&
+        obj !== null &&
+        symbol in obj &&
+        obj[symbol as keyof typeof obj] === true
+      );
+    }


The type guard implementation could be more robust. Consider strengthening the check:

Suggested change

isInstance(obj: unknown): obj is this {

return (

typeof obj === "object" &&

obj !== null &&

symbol in obj &&

obj[symbol as keyof typeof obj] === true

);

}

isInstance(obj: unknown): obj is this {

return (

typeof obj === "object" &&

obj !== null &&

symbol in obj &&

obj[symbol as keyof typeof obj] === true &&

obj instanceof Error

);

}

This ensures the object is actually an Error instance, not just any object with the symbol property.

Christian Bromann (christian-bromann) · 2026-02-11T23:52:12Z

libs/providers/langchain-google/src/chat_models/base.ts

+      statusText: response.statusText,
+    });
+
+    if (!response.ok) {


The streaming response handling could benefit from better error recovery. When parsing chunks fails, the current implementation might not provide enough context:

Suggested change

if (!response.ok) {

} catch (parseError) {

// If parsing fails, still yield a chunk with error context

const errorMessage = parseError instanceof Error ? parseError.message : 'Unknown parsing error';

yield new ChatGenerationChunk({

message: new AIMessageChunk({

content: `[Parse Error: ${errorMessage}]`,

additional_kwargs: { parseError: true }

}),

text: `[Parse Error: ${errorMessage}]`,

});

}

mmm if we do this stream consumers cant catch errors

Christian Bromann (christian-bromann) · 2026-02-11T23:52:12Z

libs/providers/langchain-google/src/converters/params.ts

+        Object.hasOwn(config, "multiSpeakerVoiceConfig"))
+    );
+  }
+


The speech configuration conversion logic should validate the voice names against known prebuilt voices to catch configuration errors early:

Suggested change

// If this is already a SpeechConfig, validate it before returning

if (isSpeechConfig(config)) {

// Add validation for voice names if needed

return config;

}

Consider adding a list of valid prebuilt voice names for validation.

we can treat this like a model string from google -- we can't know these without an extra network request locally

Christian Bromann (christian-bromann) · 2026-02-11T23:52:12Z

libs/providers/langchain-google/src/utils/gcp-auth.ts

+
+  cache.set(cacheKey, { created: now, promise });
+  return await promise.then((data) => data.token);
+}


The token caching mechanism could lead to race conditions when multiple requests are made simultaneously. Consider using a more robust caching strategy or adding proper locking mechanisms to prevent duplicate token requests.

Christian Bromann (christian-bromann) · 2026-02-11T23:52:12Z

libs/langchain-core/src/messages/block_translators/google.ts

+          return {
+            type: "tool_call",
+            id: message.id,
+            name: block.functionCall.name,
+            args: block.functionCall.args,
+          };


The tool_call conversion assumes message.id is available, but this might not always be the case. Add a fallback:

Suggested change

return {

type: "tool_call",

id: message.id,

name: block.functionCall.name,

args: block.functionCall.args,

};

return {

type: "tool_call",

id: message.id || `tool_call_${Date.now()}_${Math.random().toString(36).substr(2, 9)}`,

name: block.functionCall.name,

args: block.functionCall.args,

};

we shouldn't do this

Co-authored-by: afirstenberg <github@addventure.com>

This comment was marked as duplicate.

Sign in to view

github-actions bot added the ready label Feb 11, 2026

This comment was marked as duplicate.

Sign in to view

Christian Bromann (christian-bromann) requested changes Feb 11, 2026

View reviewed changes

Hunter Lovell (hntrl) and others added 17 commits February 11, 2026 17:09

feat: @langchain/google

5549fa4

cr

52cc5b0

test stubs

8493e40

chore(google): test + converter enhancements (#9598)

57f0100

Co-authored-by: afirstenberg <github@addventure.com>

wip

6c0ca32

cr

fb500f7

cr

f96ff6f

README

c3aeb35

cr

2478320

cr

e9304ac

chore: fix empty configs

2d7eadd

denoise test logs

e31e1d3

format

9977cd8

format better

ee58c98

skip 2.0 tool test

a8365ec

cr

bcd005b

cr

8643cd9

Hunter Lovell (hntrl) force-pushed the hunter/google branch from 1c4ba50 to 8643cd9 Compare February 12, 2026 01:09

Hunter Lovell (hntrl) added 2 commits February 11, 2026 17:09

cr

4769a62

cr

77f1689

Hunter Lovell (hntrl) requested a review from Christian Bromann (christian-bromann) February 12, 2026 01:23

Christian Bromann (christian-bromann) approved these changes Feb 12, 2026

View reviewed changes

Hunter Lovell (hntrl) merged commit 71d08c0 into main Feb 12, 2026
36 of 38 checks passed

Hunter Lovell (hntrl) deleted the hunter/google branch February 12, 2026 01:29

github-actions bot mentioned this pull request Feb 12, 2026

chore: version packages #9999

Merged

Mia (miadisabelle) pushed a commit to avadisabelle/ava-langchainjs that referenced this pull request Feb 18, 2026

feat: @langchain/google (langchain-ai#10000)

66f7f2d

Co-authored-by: afirstenberg <github@addventure.com>

picopico (picopicos) mentioned this pull request Feb 18, 2026

MCP Client Tool doesn't work with gemini models n8n-io/n8n#15553

Open

Andrew Kondratev (andruhon) mentioned this pull request Feb 23, 2026

Migrate to new @langchain/google Galvanized-Pukeko/gaunt-sloth-assistant#265

Closed

Nick Winder (nickwinder) pushed a commit to nickwinder/langchainjs that referenced this pull request Mar 24, 2026

feat: @langchain/google (langchain-ai#10000)

bb613ff

Co-authored-by: afirstenberg <github@addventure.com>

-}
+  fetch(request: Request): Promise<Response> {
+    if (this.apiKey) {
+      request.headers.set(GOOGLE_API_KEY_HEADER, this.apiKey);
+    } else if (this.credentials) {
+      request.headers.set(
+        GCP_API_KEY_HEADER,
+        `Bearer ${getGCPCredentialsAccessToken(this.credentials)}`
+      );
+    }
+    return fetch(request);
+  }

-    if (!response.ok) {
+            } catch (parseError) {
+              // If parsing fails, still yield a chunk with error context
+              const errorMessage = parseError instanceof Error ? parseError.message : 'Unknown parsing error';
+              yield new ChatGenerationChunk({
+                message: new AIMessageChunk({
+                  content: `[Parse Error: ${errorMessage}]`,
+                  additional_kwargs: { parseError: true }
+                }),
+                text: `[Parse Error: ${errorMessage}]`,
+              });
+            }

+  // If this is already a SpeechConfig, validate it before returning
+  if (isSpeechConfig(config)) {
+    // Add validation for voice names if needed
+    return config;
+  }

Conversation

Hunter Lovell (hntrl) commented Feb 11, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Why a new package?

Usage

Versioning and stability

Support policy for existing packages

Changes

New package: libs/providers/langchain-google/

@langchain/core changes

Uh oh!

changeset-bot bot commented Feb 11, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🦋 Changeset detected

Uh oh!

This comment was marked as duplicate.

Uh oh!

This comment was marked as duplicate.

Uh oh!

Christian Bromann (christian-bromann) left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Hunter Lovell (hntrl) commented Feb 11, 2026 •

edited

Loading

New package: `libs/providers/langchain-google/`

`@langchain/core` changes

changeset-bot bot commented Feb 11, 2026 •

edited

Loading