Kenji Baheux
The Chrome team is excited to see the enthusiasm around the Prompt API inChrome, part of the built-in AI effort. We announced thisexperiment at Google I/O this past May, along with the chance to sign up for theearly preview program. By signing up, youcan try early stage APIs, such as the Prompt API, and gain the opportunity toshare your feedback on how you're using these APIs in local prototypes.
It's still early days for built-in AI and the Prompt API. We hope you'llcontinue to be this excited throughout this experiment, as we change the APIand its implementation to address your feedback and make it easier to use.
Why are we running this experiment?
We have a lot to learn. What are your business needs that may benefit from AI?Are there features you want to offer your users, but can't due to prohibitivecost, privacy constraints, or latency concerns? How can we make it easier foryou to start using this technology without significant investment or deep AIknowledge?
One of the best ways to learn is through experimentation. So, we're proposingand building APIs to give you access to new, experimental capabilities offeredby on-device LLMs, such as Gemini Nano.
Chrome offers a variety of experimental APIs for developers. These APIs areaccessible with Chrome flags, which areessentially switches that enable or disable specific browser functions. Flagsare a way for us to test our hypotheses and gather feedback from developers, asyou build prototypes and experiment. The APIs may change frequently based onthis feedback, and there's no guarantee they'll ship. We may discontinue a flagif it doesn't meet expectations, or we may find we need to solve a differentproblem than initially thought.
The development of new APIs is a long process, and these experimental flags helpus learn, adapt and innovate effectively.
For built-in AI, we need your feedback to ensure we build features that havereal, practical use cases and meet your performance and quality expectations.That's why we're inviting you to join the early preview program–help us buildAI APIs you're enthusiastic to use.
What are we building?
We are building two types of APIs for built-in AI:
- Task APIs that allow developers to access built-in AI capabilities, such as atranslation API or a summarization API. Task APIs are designed to runinference against well-established models for the assignment.
- Exploratory APIs that are primarily intended for local prototyping. With theseAPIs, we intend to ask for feedback, confirm assumptions, and determine whattask APIs we build in the future. As such, exploratory APIs may never launch.
Our early preview program members can now experiment with the Prompt API, tosend natural language requests to Gemini Nano in Chrome. ThePrompt API explainerdiffers from the current implementation, as there are additional methods andenhancements we hope to implement over time.
Based on valuable feedback from early preview program members, we've learnedthat a dedicated task API might not be the best solution for every use case.Early preview participants are seeing potential for focused task APIs and themore versatile Prompt API. To explore this further, we are determining if it'spossible to make the Prompt API available in Chrome extensions.
Prompt API in Chrome Extensions
Chrome Extensions would allow you to experiment in a real environment, and wouldallow us to gain deeper insights. Based on the findings, we can refine the APIto better address real-world use cases.
Our goal is to use this simpler scope to test some ideas, learn more effectivelythan from isolated prototypes, which would ultimately support a higher qualityAPI.
This proposal is still in review, so we don't yet have a detailed timeline forwhen the Prompt API would be available in Chrome Extensions.
What's next?
We have more work to do in the early preview program, and eventually, we'llcreate origin trials to gather even morevaluable feedback from the web platform, and offer additional experimental APIs,in accordance with Chromium'slaunch process. We'recollaborating with other browser vendors, so we can work to standardize as manybuilt-in APIs as possible.
When we have a timeline, we'll share it on theChrome for Developers blog and with themailing list.Sign up to make sure you see these announcements.
There's so much innovation happening, and we're excited for the growing presenceof web-first AI. Built-in AI is just one piece of that story. Our goal is tolearn as much as possible, ensuring that we meet your needs and expectations
We understand that you may want more details. We highly recommend signing up for our early preview program and theChrome AI developer public announcements mailing list.
