Does Vertex AI Search enforce document permissions?

Yes, per user, when access control is chosen at data store creation: results are limited to documents the searching user can view, with identity from Google Identity or a federated provider such as Entra ID, Okta or Ping (checked October 2026). Hippocampus enforces folder and document grants, groups and inheritance inside the search, before ranking.

Can I keep my data in the EU?

Yes with both, with different limits. Vertex AI Search offers an EU multi-region with data residency at rest, where some features are limited compared with the global location (checked October 2026). Hippocampus stores documents and the knowledge structure in the EU (Finland); model inference runs in the US by default and can run EU-hosted or in your own environment on Corporate.

Which answering models does each support?

Vertex AI Search answers with Gemini models, chosen with a model version setting (checked October 2026). Hippocampus works with any language model: you choose which model answers, set per workspace, and on Corporate you can use your own model key.

Has Hippocampus been benchmarked against Vertex AI Search?

No: there is no published benchmark result for Vertex AI Search on our FDA corpus, so we make no accuracy comparison.

Comparison

Hippocampus vs Vertex AI Search

Q: Is Vertex AI Search being renamed Agent Search?

Yes: Google's documentation says Vertex AI Search is being renamed to Agent Search, part of the Gemini Enterprise Agent Platform, and the API is still discoveryengine.googleapis.com (checked October 2026).

18 October 2026

Last updated: October 2026

By Andreas Dahlström, Head of Deeptech Team, Co-Founder

Vertex AI Search, which Google is renaming Agent Search, part of the Gemini Enterprise Agent Platform (Google Cloud documentation, checked October 2026), is the better choice when your documents already live in Google Drive, Cloud Storage or BigQuery, when per-user permissions must come from a federated identity provider, or when your team builds on Gemini. Hippocampus is the better choice when every answer must be checked against a page and byte range, when you want to choose which language model answers, set per workspace, and when you want the knowledge base back as open Markdown. We have not benchmarked Vertex AI Search, so this article makes no accuracy comparison.

This article compares the two products dimension by dimension, and calls the product Vertex AI Search (Agent Search) below. Our earlier post Hippocampus against OpenAI File Search and PageIndex is a benchmark of other systems; Vertex AI Search was not part of it. We also compare Hippocampus vs OpenAI File Search and Hippocampus vs PageIndex.

Summary

Dimension	Hippocampus	Vertex AI Search (Agent Search)
Hosting region and self-hosting	Documents and the knowledge structure are stored in the EU (Finland), encrypted. Model inference runs in the US by default; on Corporate it can run EU-hosted or in your own environment. Own cloud or on-premises hosting is part of the Enterprise package. Security	Locations: global, plus US and EU multi-regions with data residency at rest. Google recommends global for the full feature set; some features are limited in the US and EU multi-regions. On-premises or self-hosted: not documented. Locations (checked October 2026)
Source attribution (page-level citation)	Every answer carries the document, section path, start and end page, byte offsets and the text used. Citations come from the retrieval step, not from the model.	The answer method can cite a source for each sentence. A separate Check Grounding API cites supporting chunks per claim with a support score. Page-level citation: not documented. Answer method, Check grounding (checked October 2026)
Permission enforcement at query time	Enforced inside the search as a filter before ranking; returns nothing if permissions cannot be checked. A summary over several documents is visible only to people who can read all of them. Agents obey the same permissions. Security	Results are limited to documents the searching user can view, from access lists supplied at ingestion and identity from Google Identity or federated providers (Entra ID, Okta, Ping). Covers Cloud Storage, BigQuery, Google Drive and third-party sources. Chosen at data store creation; up to 3,000 readers per document. Access control (checked October 2026)
Behaviour on unanswerable questions	When retrieval comes back empty the question is refused, not guessed. A refusal costs the same credits as an answer.	Opt-in: request options skip the answer for adversarial or non-answer-seeking queries and return search results, or return a fallback answer when content is not relevant. They default to off in the samples. Answer method (checked October 2026)
Retrieval architecture (structured memory vs vector index vs file search)	Structured memory built at import: a tree per document with an index summary, section summaries and leaf sections with pages and byte ranges. Two funnels, top-down over the summaries and bottom-up keyword plus embedding search over leaf sections, fused into one ranking. Developers	Managed parsing (digital, OCR or layout parser) and document chunking for RAG; semantic and hybrid search. Ranking and embedding internals: not documented. Parse and chunk documents, Pricing (checked October 2026)
Supported answering models	Any language model. You choose which model answers, set per workspace; on Corporate you can use your own model key or have administrators choose the models. Developers	Gemini models, chosen with a model version setting. Other models for the answer method: not documented. Answer generation models (checked October 2026)
API and MCP availability	REST API and an MCP server. The MCP server is in beta; OAuth sign-in on the shared cloud is being completed. The API on the Free tier is in beta. Developers	REST API with client libraries, a console and an embeddable widget. A Google-hosted remote MCP server with search, conversational search and engine listing tools. MCP reference (checked October 2026)
Data export format	Open Knowledge Format: a folder per document of Markdown with YAML front matter, holding the index summary, the full text, the original file and a folder per section. Readable through the API, exportable while the account is active. Developers	No export feature or format documented. Documents can be listed through the REST API and chunks read as JSON. Bulk export format: not documented. documents.list, Parse and chunk documents (checked October 2026)
Use of customer data for training	Never used to train AI, not by Thalius and not by the model providers; their contracts forbid it. Security	Google states it does not use your data to train or fine-tune models without your permission or instruction. An industry data-use addendum lets Google sample data to improve its technologies; whether it applies to a plain search app: not documented. Data governance, Data use terms (checked October 2026)
Pricing model and free tier	Credits, one unit on every tier. Free tier with monthly included credits and no card, Starter, and Corporate by quote. A refusal costs the same as an answer. Amounts on Pricing	Two models: pay-as-you-go per 1,000 queries plus index storage, or a configurable subscription with minimums. A monthly free trial allowance of queries. Amounts on the pricing page (checked October 2026)

Benchmark

No published benchmark result for Vertex AI Search on this corpus; we make no accuracy comparison. Our published benchmark uses ten public FDA documents for Saxenda, Wegovy and Zepbound (2,198 pages) and sixteen questions written by Atrogi's team. The corpus, the questions and the Hippocampus results are in the FDA GLP-1 benchmark, and every published result is on Evidence.

Where Vertex AI Search (Agent Search) is the better choice

Your documents live in Google Drive, Cloud Storage or BigQuery and permissions must follow each user's identity: Vertex AI Search (Agent Search) filters results by the searching user's access list, with identity from Google Identity or federated Entra ID, Okta or Ping (Access control, checked October 2026).
You want a separate grounding check: the Check Grounding API scores an answer candidate from 0 to 1 against supplied facts and cites supporting chunks per claim (Check grounding, checked October 2026).
You want a managed parsing pipeline inside Google Cloud: digital, OCR and layout parsers, chunking for RAG, and data residency at rest in the US or EU multi-regions (Parse and chunk documents, Locations, checked October 2026).
Your team builds on Gemini: the answer method uses Gemini models chosen with a model version setting (Answer generation models, checked October 2026).

Try Hippocampus on the same documents

Upload the documents you would put in a Vertex AI Search (Agent Search) data store and ask Hippocampus the same questions, then open the cited page for each answer. Start free, no card. Before you do, read how Hippocampus enforces permissions and where it stores documents, the API, the MCP server and the export format, how credits work on each tier and own-cloud or on-premises hosting on the Enterprise package.

Sources and last updated

Last updated: October 2026.