Google-Cloud-AI/agent-platform: a sample index, not an SDK
A curated list of code samples and tutorials for building agents on Gemini Enterprise Agent Platform (previously Vertex AI).
At a glance
- What is it?
- The repository is a curated list of notebooks and tutorials for Gemini Enterprise Agent Platform (previously Vertex AI). It ships links and a picture, not runnable code you can pip install, and that shapes who should clone it.
- Who is it for?
- Adopt this repository if you are evaluating Gemini Enterprise Agent Platform and want one page that points at ADK, Agent Runtime, Agent Evaluation and the protocol specs, then follow those links into GoogleCloudPlatform/generative-ai where the actual notebooks live.
- Can I use it commercially?
- Yes. Apache-2.0 is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
- Is it still maintained?
- Yes. The repository last received commits 114 days ago.
- What is it written in?
- GitHub does not report a main language for this repository.
Answers come from the project's GitHub data, last synced on September 18, 2026, and from our analysis. They are not legal advice.
DEEP OPEN-SOURCE ANALYSIS
What this repository actually is, and who it is for
The README opens with one sentence: a curated list of code samples and tutorials for building agents on Gemini Enterprise Agent Platform, previously Vertex AI. That is the whole product. The top level of the repository contains .gitignore, AgentPlatform.png, CODE_OF_CONDUCT.md, CONTRIBUTING.md, LICENSE, README.md and SECURITY.md. There is no source directory, no package manifest, no build file.
So the audience is narrow and specific. It is for an engineer or architect who has decided to look at Google's agent stack and wants a map of the pieces before committing time to any single one. The README groups those pieces into Build, Scale, Govern and Optimize, with a Foundation Models list above them. If you already know you want the Agent Development Kit, you do not need this page; you need adk.dev. If you are still working out whether you need ADK, Agent Runtime, Agent Garden or all three, the grouping is the value.
It is not documentation for the platform itself. The README points at docs.cloud.google.com/gemini-enterprise-agent-platform for that, plus an onboarding guide at goo.gle/agent-platform-onboard and an intro video. Treat the repository as a table of contents with commentary, and read it that way.
How the index is organised: four stages and a protocol layer
The structure follows an agent's lifecycle rather than a feature list. Build covers the development surface: Agent Development Kit, described as an open-source, model-agnostic framework for building production agents, and Agents CLI for scaffolding, testing and deploying. Alongside them sit Agent Studio for low-code prototyping and Agent Garden for pre-built templates. Tools and data integration is a separate cluster: Grounding, RAG Engine, Agent Search (previously Vertex AI Search, Generative AI App Builder), and Vector Search 2.0.
Scale covers hosting and state: Agent Runtime (previously Agent Engine, Reasoning Engine), Agent Sessions for multi-turn state, Agent Memory Bank for long-term semantic memory, and Agent Sandbox for isolated code execution. Govern collects Agent Gateway, Agent Identity, Agent Policies, Agent Registry and Model Armor. Optimize holds Agent Evaluation, Agent Simulation, Agent Observability and Agent Optimizer.
Between Build and Scale the README inserts a protocols section: A2A for multi-agent coordination and handoffs, MCP for connecting models to external tools, A2UI for agent-driven interface generation, AP2 for automated financial transactions, and UCP for commerce and retail operations. That placement is a design statement. The protocols sit under Build because they are interoperability surfaces an agent author wires up, not runtime infrastructure. Note that two Govern entries, Agent Gateway and Agent Policies, are marked Private Preview in the README, which means you cannot plan production dependencies around them yet.
Installing it: the honest answer is that there is nothing to install
Because the repository is an index, there is no install step. Cloning gives you the README and the licence files. The README does not document a package, a pip target, or a container image for this repository itself.
What you can do is clone it and read the index locally, which is useful mainly if you want the link list in your editor:
git clone https://github.com/Google-Cloud-AI/agent-platform.git
cd agent-platform
lsThe listing should show the files named above, including README.md and LICENSE. Nothing executable appears.
The first real use is therefore to pick one linked target and follow it. The README's onboarding entry is a redirect short link, and the actual sample notebooks it points to live in the GoogleCloudPlatform/generative-ai repository, for example the Agent Engine intro notebook and the Memory Bank getting-started notebook. A sensible first move is to open the Agent Development Kit site at adk.dev, since the README calls it the framework for building production agents, and only then return to the index to decide what hosting, evaluation and governance pieces you need around it. If you want a hands-on path rather than a reading list, go straight to the generative-ai notebooks and skip this repository entirely.
Where the index breaks down
The main limitation is that a curated list ages faster than the software it describes. The last push to this repository was on 2026-05-27, and the README itself carries rename notes in several places: Agent Runtime was previously Agent Engine and Reasoning Engine, Agent Search was previously Vertex AI Search and Generative AI App Builder, and the platform was previously Vertex AI. Renames mean search results, blog posts and Stack Overflow answers you find will use different names for the same thing, and the index cannot reconcile that for you.
A second issue is that the README mixes maturity levels without a consistent marker. Agent Gateway and Agent Policies are labelled Private Preview, but the other Govern entries carry no status label at all, so you cannot tell from this page which are generally available and which are not. The README also does not document rollback, deprecation timelines or migration paths between the old and new names.
Finally, the repository is the wrong tool if you want to evaluate behaviour. There are no benchmarks here, no cost figures, and no comparison of when to choose RAG Engine over Agent Search over Vector Search 2.0 for a given retrieval problem. Each is one line of description. Choosing among them requires the linked docs, not this index.
The realistic alternative: go straight to the sample repository
The obvious alternative is GoogleCloudPlatform/generative-ai, which this README links to repeatedly for the actual notebooks: grounding, RAG Engine, Vector Search 2.0, Agent Runtime, Memory Bank, code execution, and Agent Evaluation. The difference in approach is concrete. This repository is a hand-written index with descriptions and grouping; generative-ai is a large notebook collection where the code lives. If you learn by running cells, generative-ai is the destination and this repository is at best a signpost to it.
A second alternative is the platform documentation site itself, which the README names as the homepage. That is the authoritative source for API surface and current naming, and it is where you should check anything the index asserts.
There is also a genuine architectural fork inside the index that matters more than any external comparison: the README lists both Agent Development Kit, an open-source model-agnostic framework, and Agent Studio, a low-code visual interface. Those represent different build philosophies for the same platform, and the index presents them as peers without saying when code-first beats visual prototyping. That decision is yours to make from the linked docs.
Maintenance, licence and what adopting it costs you
The repository is not archived, and the last push was on 2026-05-27. That is roughly four months before the date of writing, so it is not abandoned, but it is also not something updated on a weekly cadence, and the README's own rename notes show that the underlying platform moves faster than the list does. Your upgrade cost is not a dependency bump; it is re-reading the index when platform names change and re-checking each linked target, because the links point at external sites and repositories that this project does not control. A dead or moved link is a silent failure here.
The licence is Apache-2.0, per the LICENSE file at the top level. For a link index and image asset that is permissive and unremarkable, but note that the licence covers this repository's contents, not the linked samples, notebooks or platform services. Those carry their own terms, and the Google Cloud services referenced throughout are commercial products with their own pricing and preview conditions. Nothing here changes that. This is a description of the licence as stated in the repository, not legal advice.
Editorial conclusion
Adopt this repository if you are evaluating Gemini Enterprise Agent Platform and want one page that points at ADK, Agent Runtime, Agent Evaluation and the protocol specs, then follow those links into GoogleCloudPlatform/generative-ai where the actual notebooks live. Do not adopt it if you expected an installable SDK, a CLI, or code you can run after cloning: the top level holds only .gitignore, AgentPlatform.png, CODE_OF_CONDUCT.md, CONTRIBUTING.md, LICENSE, README.md and SECURITY.md. Before relying on anything here, open the specific linked notebook and confirm the API surface still matches, because the last push was on 2026-05-27 and the platform renamed Vertex AI to Gemini Enterprise Agent Platform, which means older tutorials and third-party write-ups may use the previous names for the same services.
Frequently asked questions
What is Google's agent platform?
In this repository it is Gemini Enterprise Agent Platform, previously Vertex AI, described as the platform on which the listed samples and tutorials build agents. The README points to docs.cloud.google.com/gemini-enterprise-agent-platform as the documentation home.
What is the purpose of the agent Platform API?
The repository does not describe an API surface itself. It indexes the pieces you would call, including Agent Development Kit for building agents, Agent Runtime for hosting them, and Agent Sessions for multi-turn state, and links to the platform documentation for the actual interfaces.
How do I use Gemini Enterprise Agent Platform?
The README gives an onboarding guide and an intro video as starting points, then lists build, scale, govern and optimize components. The runnable notebooks it references live in the GoogleCloudPlatform/generative-ai repository rather than in this one.
What is an agent platform?
The repository does not define the term in general. It shows one instance of it: a platform whose listed parts cover building agents, hosting and scaling them, governing their identities and policies, and evaluating and optimizing their behaviour.
What is agent platform in GCP?
The README places the platform inside Google Cloud and links to docs.cloud.google.com for it, with components such as Agent Runtime, Agent Sessions and Model Armor. Several entries note earlier names, including Agent Runtime previously Agent Engine and Reasoning Engine.
Community notes