Is PageIndex open source?

Yes: PageIndex's local mode is MIT licensed and runs on your own machine with your own LLM key (checked October 2026). PageIndex Cloud and its Enterprise deployments are hosted or sales-led offers. Hippocampus runs on Thalius's EU cloud, or in your own cloud or on-premises as part of the Enterprise package.

Does Hippocampus use embeddings, and does PageIndex?

Hippocampus does, alongside structure: it builds a tree per document and fuses a top-down path over the summaries with keyword plus embedding search over leaf sections. PageIndex describes itself as vectorless: an LLM reasons over a tree index per document (checked October 2026).

Which answering models can I use?

Both work with more than one provider. PageIndex works with OpenAI, Anthropic, OpenRouter or any OpenAI-compatible endpoint, and LLM inference is billed separately (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.

Which one gives finer citations?

It depends on the mode: PageIndex Cloud cites at block level with an optional bounding box, and local mode at page level (checked October 2026). Hippocampus cites the document, section path, start and end page and byte offsets, with the text used.

Was PageIndex's own service benchmarked?

No: our FDA benchmark used our adaptation of PageIndex's published tree-search approach, not PageIndex's own retrieval and chat services. That setup scored 33 of 48 against 40 of 48 for Hippocampus, with GPT-5.4 mini answering for every system on 17 September 2026.

Comparison

Hippocampus vs PageIndex

18 October 2026

Last updated: October 2026

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

Hippocampus is the better choice when you want a hosted memory layer with EU storage, permissions enforced inside the search, and citations down to the page and byte range. PageIndex is the better choice when you want MIT-licensed code on your own machine with your own LLM key, indexing priced per page with no retrieval fee, or block-level citations on PageIndex Cloud. On our FDA benchmark, with GPT-5.4 mini answering for every system, Hippocampus answered 40 of 48 questions correctly and our adaptation of the PageIndex approach 33 of 48*; PageIndex's own services were not used. The PageIndex setup was faster: a median of 8 seconds against 20.

This article compares the two products dimension by dimension. The benchmark run itself, with five systems side by side and the questions where they differ, is our earlier post Hippocampus against OpenAI File Search and PageIndex; this article does not repeat it. We also compare Hippocampus vs OpenAI File Search and Hippocampus vs Vertex AI Search.

Summary

Dimension	Hippocampus	PageIndex
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	Open-source local mode (MIT) runs on your own machine with your own LLM key. Enterprise lists SaaS, dedicated deployment, private VPC on AWS, Azure or GCP, and on-premises including air-gapped. Hosted region names: not documented. GitHub, Enterprise (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.	Local mode cites at page level; Cloud adds block-level citations with an optional bounding box. Chat SDK, Agents SDK (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	Per-user document permissions: not documented. Enterprise lists role-based access control, SSO, multi-user workspaces and audit logs; the agent docs say framework tools keep access to the whole library. Enterprise, Agents SDK (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.	The open-source agent prompt tells the model to say so when the documents do not answer. This is a prompt instruction; the docs do not describe it. agent_tools.py (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	Vectorless: a hierarchical tree index per document, and an LLM reasons over the tree to find and read sections. GitHub (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	Any provider: OpenAI, Anthropic, OpenRouter or any OpenAI-compatible endpoint. Chat SDK, Getting started (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	Python and JavaScript SDKs. A hosted MCP server, Cloud only, for documents already indexed in Cloud. REST endpoint reference: not documented. MCP, Getting started (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	Per document through the SDK: the tree as JSON and the extracted text as Markdown. Whole-account export: not documented. Documents SDK, Terms (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	Training on customer data: not documented. The Terms let the supplier use customer data to improve the software. Terms, Enterprise (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	Cloud is usage-based with no monthly minimum: $0.01 per page indexed once, $0.001 per active page per month with the first 1,000 free, no retrieval fees, LLM inference not included. $10 in free credits, no card. Local mode is free; Enterprise is sales-led. Pricing (checked October 2026)

Benchmark

Ten public FDA documents for Saxenda, Wegovy and Zepbound (2,198 pages), sixteen questions written by Atrogi's team, each asked three times in independent conversations. Every system answered through the same model, GPT-5.4 mini, on 17 September 2026. A grading model, GPT-5 with three votes and a majority verdict, compared each answer with a reference answer; declining the control question counts as correct.

Hippocampus: 40 of 48 correct*, 79% weighted, median 20 seconds per answer.
PageIndex approach, our adaptation: 33 of 48 correct*, 65% weighted, median 8 seconds per answer.

The PageIndex row is our adaptation of their published tree-search approach: one step, up to six sections, 12,000 characters per section. PageIndex's own retrieval and chat services were not used. Those limits can disadvantage answers spread across many sections or sitting in long tables.

The weighted score counts easy questions once, medium twice and hard three times. On the control question, which the documents cannot answer, the PageIndex setup declined every time; Hippocampus said each time that the documents cannot support a ranking, and in two of three answers offered the approval order as a labelled proxy.

What this result does not show:

The gap between the top two systems in the full table, Hippocampus at 40 and OpenAI File Search at 36 on average, is within run-to-run variation.
Hippocampus configurations tested on 16 to 17 September scored between 37 and 40 of 48. The figure above is a run that scored 40.
The search guide Hippocampus gives the model was refined on these same sixteen questions. The other systems ran on vendor defaults plus a one-sentence instruction to answer only from the material.
File Search ran twice. The other systems, PageIndex included, ran once, so differences of a few answers can come from variation alone.
Hippocampus is the slowest system tested: a median of 20 seconds per answer against 5 to 8 seconds for the others.
The grading model is OpenAI's GPT-5, and the answering model and two of the systems tested are also OpenAI's.

The full table with all five systems is in Hippocampus against OpenAI File Search and PageIndex. The corpus, the questions and the results with three answering models are in the FDA GLP-1 benchmark, and every published result is on Evidence.

* Answering model GPT-5.4 mini for every system. Grading by GPT-5, three votes, majority verdict. Thalius evaluation, 17 September 2026.

Where PageIndex is the better choice

You want open-source code you control: local mode is MIT licensed, runs on your own machine with your own LLM key, and Enterprise adds private VPC and air-gapped on-premises options (GitHub, Enterprise, checked October 2026).
You want to pay for indexing, not for retrieval: Cloud charges per page indexed and per active page, with no retrieval fees, and you bring any OpenAI-compatible LLM (Pricing, Chat SDK, checked October 2026).
You need citations that point inside the page: Cloud returns block-level citations with an optional bounding box, and offers a hosted MCP server and SDKs for agent frameworks (Agents SDK, MCP, checked October 2026).
Speed matters more than depth: in our FDA benchmark the PageIndex setup answered in a median of 8 seconds against 20 for Hippocampus, and declined the control question every time.

Try Hippocampus on the same documents

Upload the documents you would index in PageIndex and ask Hippocampus the same questions, then compare the citations side by side. 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.