Add First-Class Gemini SDK Integration to Contrib

[JasonSteving99]

Temporal Integration for the Google Gemini SDK

This adds a first-class integration that lets users call the Gemini SDK's AsyncClient directly from within Temporal workflows. Every API call and tool invocation becomes a durable Temporal activity — giving full crash recovery, visibility in workflow event history, and replay safety — while keeping credentials entirely on the worker side.

How it works

The integration shims three layers of the Gemini SDK so that workflows can use client.models, client.files, client.file_search_stores, client.chats, and all other SDK modules naturally:

TemporalApiClient (_temporal_api_client.py)

A BaseApiClient subclass that replaces the SDK's HTTP layer. Instead of making network calls, async_request and async_request_streamed serialize the request and dispatch it through workflow.execute_activity. The real HTTP call happens inside the activity on the worker, where the actual genai.Client with real credentials lives. Sync methods raise immediately. Per-request http_options are validated (non-serializable fields like httpx_client are rejected), and timeout is mapped to Temporal's start_to_close_timeout.

TemporalAsyncFiles / TemporalAsyncFileSearchStores (_temporal_files.py, _temporal_file_search_stores.py)

Subclasses of AsyncFiles and AsyncFileSearchStores that override upload, download, register_files, and upload_to_file_search_store to dispatch the entire operation as a Temporal activity. This avoids filesystem access (os module) and credential token refresh in the workflow sandbox. Methods like get, delete, list are inherited and work through the TemporalApiClient's async_request activity. File uploads accept str paths (resolved on the worker), os.PathLike, or io.IOBase (bytes serialized across the activity boundary).

TemporalAsyncClient (_temporal_async_client.py)

An AsyncClient subclass that wires in TemporalAsyncFiles and TemporalAsyncFileSearchStores. All other SDK modules (models, tunings, caches, batches, live, tokens, operations) are inherited unchanged since they only use async_request under the hood.

GeminiPlugin (_gemini_plugin.py)

A SimplePlugin that registers all activities, configures the Pydantic data converter, and passes google.genai through the workflow sandbox. Users pass a fully configured genai.Client — the plugin never constructs one itself. An optional extra_credentials parameter supports operations like register_files that need separate GCS credentials.

activity_as_tool (workflow.py)

Wraps any @activity.defn function so it looks like a regular async callable to Gemini's automatic function calling (AFC). When the model decides to call the tool, the SDK invokes the wrapper, which dispatches through workflow.execute_activity. Users can also pass plain workflow methods directly as tools — these run in-workflow without an activity.

Batched streaming

generate_content_stream is supported via a batched approach: the async_request_streamed activity collects all chunks from the real streaming response and returns them as a list. The workflow-side TemporalApiClient yields them back as an async generator so the SDK sees the expected interface.

Usage

# Worker side
client = genai.Client(api_key=os.environ["GOOGLE_API_KEY"])
plugin = GeminiPlugin(client)

# Workflow side
@workflow.defn
class MyWorkflow:
    @workflow.run
    async def run(self, query: str) -> str:
        client = gemini_client()
        response = await client.models.generate_content(
            model="gemini-2.5-flash",
            contents=query,
            config=types.GenerateContentConfig(
                tools=[activity_as_tool(my_tool)],
            ),
        )
        return response.text

Testing

31 integration tests covering:

• Basic generate_content and multi-chunk streaming
• AFC tool calling (single-arg, multi-arg, workflow methods, sequential multi-tool, failure propagation)
• Per-request http_options propagation (headers, api_version, base_url)
• File upload via str path and io.BytesIO, file download
• File search store upload
• Multi-turn chat via client.chats
• TemporalAsyncClient wiring verification
• TemporalApiClient error paths (sync raises, low-level upload/download raises)
• activity_as_tool validation and signature preservation
• A full end-to-end integration test that exercises all real activity implementations (generate, stream, file upload, download, store upload, RAG query, store delete) with a mocked genai.Client — ensuring the actual activity code in _gemini_activity.py is covered, not just the workflow-side shims.

[CLAassistant]

All committers have signed the CLA.

[semgrep-managed-scans]

Semgrep found 1 ssc-80bc2470-4c90-4cf6-811e-5a265b8f01d5 finding:

• uv.lock
  • L2102 - Triage

Risk: Affected versions of litellm are vulnerable to Exposure of Sensitive Information to an Unauthorized Actor / Use of Password Hash With Insufficient Computational Effort / Use of a Broken or Risky Cryptographic Algorithm. LiteLLM exposes password hashes via authenticated API endpoints and its login flow accepts a stored SHA-256 hash as a valid password, allowing an authenticated user to steal another user's hash and log in as that user, resulting in full authentication bypass and privilege escalation.

Fix: Upgrade this library to at least version 1.83.0 at sdk-python/uv.lock:2102.

Reference(s): GHSA-69x8-hrgq-fjj8

Semgrep found 1 ssc-9ebb5c4b-d829-4496-b42b-76424d99a7b1 finding:

• uv.lock
  • L2102 - Triage

Risk: Affected versions of litellm are vulnerable to Incorrect Authorization. This vulnerability stems from inadequate enforcement of access control policies, allowing authenticated users to perform actions beyond their intended privilege level and potentially alter sensitive system configurations or access restricted resources.

Fix: Upgrade this library to at least version 1.83.0 at sdk-python/uv.lock:2102.

Reference(s): GHSA-53mr-6c8q-9789, CVE-2026-35029

Semgrep found 1 ssc-4464d88f-a369-4669-9a83-d26a12248ef5 finding:

• uv.lock
  • L982 - Triage

Risk: Affected versions of google-adk are vulnerable to Missing Authentication for Critical Function. Google Agent Development Kit (ADK) contains a missing authentication flaw that can let an unauthenticated remote attacker reach code-injection paths and execute arbitrary code on the server running the ADK instance, including ADK Web deployments on Python, Cloud Run, and GKE.

Fix: Upgrade this library to at least version 1.28.1 at sdk-python/uv.lock:982.

Reference(s): GHSA-rg7c-g689-fr3x, CVE-2026-4810