Add JS/TS cookbooks for Google Gemini tracing (basic + multimodal)

[JK-Amanda-Goo]

Summary

Adds two JS/TS cookbooks for native Google Gemini tracing with Langfuse (closes #2822). Both examples come from a working Next.js application.

js_integration_google_gemini.ipynb — basic text generation tracing:

• Wraps generateContent calls with langfuse.trace() + trace.generation()
• Passes usageMetadata fields (promptTokenCount, candidatesTokenCount, totalTokenCount) for accurate cost tracking
• Documents the correct model name string (e.g. "gemini-2.0-flash") needed to match Langfuse's model registry
• Shows the streaming pattern (generateContentStream) with token capture after stream completion
• Includes the flushAsync() pattern for serverless/Next.js with a complete Route Handler example

js_integration_google_gemini_multimodal.ipynb — image + text tracing:

• Passes inline base64 images to Gemini using inlineData
• Logs images in the OpenAI message format (image_url content block) so they render in the Langfuse UI
• Explains that promptTokenCount covers combined image + text tokens
• Includes a complete Next.js Route Handler example with flushAsync()

Notes

There is currently no @arizeai/openinference-instrumentation-google-generative-ai package for JS (unlike Python), so these cookbooks use the Langfuse JS SDK for manual tracing. Happy to adjust the approach if there is a preferred pattern for manual instrumentation in the JS cookbook collection.

Test plan

• Both notebooks run end-to-end with a valid GEMINI_API_KEY and Langfuse keys
• Traces appear in Langfuse UI with correct token counts and model cost
• Images render inline in the multimodal trace
• flushAsync() examples confirmed working in a Next.js serverless environment

🤖 Generated with Claude Code

Disclaimer: Experimental PR review

Greptile Summary

This PR adds two new JS/TS cookbooks demonstrating manual Langfuse tracing for Google Gemini — one for basic text generation (including streaming) and one for multimodal image+text calls. The overall approach (manual trace/generation wrapping, usageMetadata token capture, flushAsync() for serverless) is sound and fills a genuine gap since no auto-instrumentation exists for the Gemini JS SDK.

• The streaming cell in js_integration_google_gemini.ipynb uses process.stdout.write, which is a Node.js API unavailable in the Deno kernel — this throws ReferenceError at runtime.
• The same notebook re-declares const genAI, const MODEL, and const prompt in the streaming cell after they are already declared in Step 4, causing SyntaxError: Identifier 'MODEL' has already been declared when cells are run sequentially in Deno's shared REPL scope.

Confidence Score: 3/5

Not safe to merge as-is — two P1 runtime errors in the streaming cell will cause the notebook to fail when executed in Deno.

Two P1 bugs in the basic cookbook (wrong stdout API for Deno, const re-declarations) mean the streaming example cannot run as written. The multimodal cookbook is cleaner but has a P2 reliability concern for large images. Both issues are straightforward to fix.

cookbook/js_integration_google_gemini.ipynb — streaming cell needs process.stdout.write replaced and duplicate const declarations removed.

Important Files Changed

Filename	Overview
cookbook/js_integration_google_gemini.ipynb	New cookbook for basic Gemini JS/TS tracing; two runtime bugs: process.stdout.write (Node.js API in a Deno notebook) and const re-declarations between cells that will throw SyntaxErrors when run sequentially.
cookbook/js_integration_google_gemini_multimodal.ipynb	New cookbook for multimodal Gemini tracing; spread-based base64 conversion can stack-overflow for large images, but the overall tracing logic and image rendering approach are correct.

Sequence Diagram

sequenceDiagram
    participant App as App Code (JS/TS)
    participant LF as Langfuse SDK
    participant Gemini as Google Gemini API
    participant LFBE as Langfuse Backend

    App->>LF: langfuse.trace({ name, input })
    App->>LF: trace.generation({ name, model, input })
    App->>Gemini: model.generateContent(prompt)
    Gemini-->>App: result (text + usageMetadata)
    App->>LF: generation.end({ output, usage })
    App->>LF: trace.update({ output })
    App->>LF: await langfuse.flushAsync()
    LF->>LFBE: Batch POST events (trace + generation)
    LFBE-->>LF: 200 OK

Loading

Prompt To Fix All With AI

This is a comment left during a code review.
Path: cookbook/js_integration_google_gemini.ipynb
Line: 192

Comment:
**`process` is not defined in Deno**

`process.stdout.write` is a Node.js API. The notebook uses the Deno kernel (all other cells use `Deno.env`), where `process` is not a global — this line will throw `ReferenceError: process is not defined` at runtime. Replace with `Deno.stdout.write(new TextEncoder().encode(part))`.

```suggestion
  Deno.stdout.write(new TextEncoder().encode(part));
```

How can I resolve this? If you propose a fix, please make it concise.

---

This is a comment left during a code review.
Path: cookbook/js_integration_google_gemini.ipynb
Line: 174-177

Comment:
**`const` re-declarations clash with Step 4 cell in Deno REPL**

`const genAI`, `const MODEL`, and `const prompt` are already declared in the Step 4 cell (lines 105–108). Deno's Jupyter kernel shares a single REPL scope across cells, so running this streaming cell after Step 4 will throw `SyntaxError: Identifier 'MODEL' has already been declared`. Consider renaming these to unique identifiers (e.g. `streamGenAI`, `streamModel`, `streamPrompt`) or removing the duplicate declarations.

How can I resolve this? If you propose a fix, please make it concise.

---

This is a comment left during a code review.
Path: cookbook/js_integration_google_gemini_multimodal.ipynb
Line: 115

Comment:
**Spread into `String.fromCharCode` fails for large images**

Spreading a `Uint8Array` as function arguments copies all bytes onto the call stack. For images larger than ~100–200 KB this will throw `RangeError: Maximum call stack size exceeded`. Users adapting this for real photos will hit this silently. Use `.map()` instead.

```suggestion
const imageBase64 = btoa([...new Uint8Array(imageBuffer)].map(b => String.fromCharCode(b)).join(""));
```

How can I resolve this? If you propose a fix, please make it concise.

_{Reviews (1): Last reviewed commit: "Add JS/TS cookbooks for Google Gemini tr..." | Re-trigger Greptile}

Greptile also left 2 inline comments on this PR.

[claude]

Claude Code Review

This pull request is from a fork — automated review is disabled. A repository maintainer can comment @claude review to run a one-time review.

[vercel]

@JK-Amanda-Goo is attempting to deploy a commit to the langfuse Team on Vercel.

A member of the Team first needs to authorize it.

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[greptile-apps]

process is not defined in Deno

process.stdout.write is a Node.js API. The notebook uses the Deno kernel (all other cells use Deno.env), where process is not a global — this line will throw ReferenceError: process is not defined at runtime. Replace with Deno.stdout.write(new TextEncoder().encode(part)).

Suggested change

	" process.stdout.write(part);\n",
	Deno.stdout.write(new TextEncoder().encode(part));

Prompt To Fix With AI

This is a comment left during a code review.
Path: cookbook/js_integration_google_gemini.ipynb
Line: 192

Comment:
**`process` is not defined in Deno**

`process.stdout.write` is a Node.js API. The notebook uses the Deno kernel (all other cells use `Deno.env`), where `process` is not a global — this line will throw `ReferenceError: process is not defined` at runtime. Replace with `Deno.stdout.write(new TextEncoder().encode(part))`.

```suggestion
  Deno.stdout.write(new TextEncoder().encode(part));
```

How can I resolve this? If you propose a fix, please make it concise.

[greptile-apps]

const re-declarations clash with Step 4 cell in Deno REPL

const genAI, const MODEL, and const prompt are already declared in the Step 4 cell (lines 105–108). Deno's Jupyter kernel shares a single REPL scope across cells, so running this streaming cell after Step 4 will throw SyntaxError: Identifier 'MODEL' has already been declared. Consider renaming these to unique identifiers (e.g. streamGenAI, streamModel, streamPrompt) or removing the duplicate declarations.

Prompt To Fix With AI

This is a comment left during a code review.
Path: cookbook/js_integration_google_gemini.ipynb
Line: 174-177

Comment:
**`const` re-declarations clash with Step 4 cell in Deno REPL**

`const genAI`, `const MODEL`, and `const prompt` are already declared in the Step 4 cell (lines 105–108). Deno's Jupyter kernel shares a single REPL scope across cells, so running this streaming cell after Step 4 will throw `SyntaxError: Identifier 'MODEL' has already been declared`. Consider renaming these to unique identifiers (e.g. `streamGenAI`, `streamModel`, `streamPrompt`) or removing the duplicate declarations.

How can I resolve this? If you propose a fix, please make it concise.