Gemini API Now Supports Event-Driven Webhooks for Long-Running AI Jobs

If you've ever managed an AI pipeline that runs lengthy tasks—like processing thousands of prompts overnight, launching a Deep Research agent, or generating a long video—you've likely experienced the frustration of polling. Your code constantly sends GET requests asking, 'Are you done yet?' This wastes resources, adds latency, and becomes a reliability nightmare at scale. Google has addressed this with a new event-driven webhook system for the Gemini API, moving from inefficient polling to push-based notifications.

What problem does the new webhook feature solve?

Polling for long-running operations (LROs) in AI workflows is inefficient. Before webhooks, developers had to repeatedly call GET /operations to check if a job completed. As Gemini handles more agentic and high-volume tasks—like Deep Research, batch prompt processing, or video generation—operations can take minutes or hours. Polling for hours burns compute resources and API quota, and introduces delays between job completion and notification. This new feature replaces that with a push-based notification: the Gemini API sends an HTTP POST to your server the moment a task finishes, eliminating the need for constant status checks.

Gemini API Now Supports Event-Driven Webhooks for Long-Running AI Jobs

How do webhooks work in the Gemini API?

Event-driven webhooks provide a push-based system. When a long-running operation completes, the Gemini API immediately sends a real-time HTTP POST payload to a configured endpoint on your server. Instead of your code asking 'are you done?', the API proactively notifies you. This reduces latency and overhead. There are two configuration modes: static and dynamic webhooks, each suited for different use cases. The feature is available for all developers using the Gemini API and targets a core pain point in agentic and high-volume AI workflows.

What are the benefits of using webhooks over polling?

Webhooks offer several advantages. First, they eliminate wasteful polling loops, saving compute resources and API quota. Second, they reduce latency: your application learns about completion almost instantly, rather than waiting for the next polling interval. Third, they improve reliability by removing the risk of missed status updates or rate limits. At scale, this becomes critical for agentic workflows, batch processing, and long-running tasks like video generation. Webhooks also simplify code by removing the need for complex polling logic and retry mechanisms.

What are the two configuration modes for webhooks?

The Gemini API supports static and dynamic webhooks. Static webhooks are project-level endpoints configured via the WebhookService API. They are registered once per project and trigger for any matching event, making them ideal for global integrations like notifying Slack or syncing a database. Dynamic webhooks are request-level overrides: you pass a webhook URL in the webhook_config payload when making a specific job call. This allows routing individual jobs to dedicated endpoints, perfect for agent-orchestration queues where different tasks need different handlers.

How do static webhooks differ from dynamic webhooks?

Think of static webhooks as a standing instruction: 'Always deliver packages to the front desk.' They apply globally to all operations in a project. Dynamic webhooks are like saying: 'For this one shipment, send it to my home address.' They are specified per request. An additional dynamic webhook feature is the user_metadata field, which lets you attach arbitrary key-value metadata (e.g., {"job_group": "nightly-eval", "priority": "high"}) when dispatching a job. This metadata travels with the notification, enabling smarter routing and processing logic in your application.

What is the user_metadata field and how is it used?

The user_metadata field is a feature of dynamic webhooks. It allows you to attach custom key-value pairs to a job at dispatch time. For example, you could add {"job_group": "nightly-eval", "priority": "high"}. This metadata is included in the webhook payload when the job completes, so your server can process it accordingly. It is particularly useful when you have multiple types of jobs hitting the same endpoint; you can use the metadata to route, prioritize, or log the task without needing to query the API again.

How does this feature impact AI workflows?

This eliminates a major bottleneck in production AI pipelines. For agentic workflows where tasks are triggered asynchronously, webhooks enable faster, more reliable orchestration. High-volume batch processing becomes more efficient by freeing up resources. Long-running tasks like video generation or Deep Research no longer require polling loops that consume quota. The feature scales seamlessly, allowing developers to build responsive, event-driven systems. It marks a shift from 'ask and wait' to 'listen and react,' aligning with modern event-driven architecture patterns.