Is Uthereal Cortex a LlamaIndex alternative?

For document question answering, yes. LlamaIndex is a data framework: ingestion, node parsing, indexing, query engines and evaluation, run by you. @uthereal-sdk/cortex is a hosted service returning cited answers, so there is no index to build or maintain.

LlamaIndex returns source nodes. Is that the same as citations?

No. Source nodes tell you which chunks were retrieved. A citation ties one claim in the answer to one passage, with the document, page and span, so a reader can verify that specific statement. The second is what regulated users ask for.

Which is better for a large, changing corpus?

With LlamaIndex you own the re-ingestion pipeline and its cost every time chunking or the embedding model changes. With Cortex, uploading or removing a document is the whole operation.

Can I keep LlamaIndex and use Cortex for retrieval?

Yes. Call the rag endpoint for ranked JSON evidence and use it as a retriever inside your existing query engine.

Is TypeScript a constraint?

The SDK is TypeScript-first and ESM, running on Node.js 22+, Deno 2, Bun, Cloudflare Workers and browsers. LlamaIndex is strongest in Python. For a TypeScript product, the SDK avoids a second runtime in your stack.

Uthereal Cortex SDK vs LlamaIndex

LlamaIndex is an excellent toolkit for people whose job is retrieval. Most teams shipping a knowledge product do not want that job; they want answers their users can check.

Last updated 18 September 2026 · SDK v1.1.0 · changelog

Short answer: choose LlamaIndex when you want to own ingestion, indexing and evaluation, typically in Python. Choose @uthereal-sdk/cortex when you want cited answers in a TypeScript product without operating an index, with claim-level references and highlighted PDF pages returned by the API.

Create your agent Read the SDK guide

Different scopes

LlamaIndex - connectors, node parsers, indexes, query engines, evaluators. Deep control over retrieval.
Uthereal Cortex SDK - a hosted retrieval and answer service with React components for citations and PDF evidence.
Both answer questions over documents. Only one of them hands you the index to operate.

Source nodes versus claim-level citations

A list of retrieved nodes shows what the system looked at. It does not show which sentence supports which statement. Cortex returns the answer decomposed into claims, each bound to a reference with a document, page and span, and resolves that reference to a rendered page with the passage highlighted.

That mapping is why an editor, auditor or clinician will accept the output. It is also the part that is hardest to retrofit onto a home-grown pipeline.

What operations look like

Ingestion - LlamaIndex: your pipeline and schedule. Cortex: upload to the agent.
Re-indexing - LlamaIndex: on every chunking or embedding change. Cortex: none.
Evaluation - LlamaIndex: you build and maintain the eval set. Cortex: retrieval quality is the service's responsibility.
Residency - LlamaIndex: wherever you host it. Cortex: EU or Swiss resident, zero retention, no training on your content.

How to choose

Pick LlamaIndex for research, custom retrieval strategies and Python-centred stacks
Pick Cortex for TypeScript products where provenance is a requirement and time to launch matters
Use the rag endpoint to keep your query engine and swap only the evidence source

Frequently asked questions

Is Uthereal Cortex a LlamaIndex alternative?: For document question answering, yes. LlamaIndex is a data framework: ingestion, node parsing, indexing, query engines and evaluation, run by you. @uthereal-sdk/cortex is a hosted service returning cited answers, so there is no index to build or maintain.
LlamaIndex returns source nodes. Is that the same as citations?: No. Source nodes tell you which chunks were retrieved. A citation ties one claim in the answer to one passage, with the document, page and span, so a reader can verify that specific statement. The second is what regulated users ask for.
Which is better for a large, changing corpus?: With LlamaIndex you own the re-ingestion pipeline and its cost every time chunking or the embedding model changes. With Cortex, uploading or removing a document is the whole operation.
Can I keep LlamaIndex and use Cortex for retrieval?: Yes. Call the rag endpoint for ranked JSON evidence and use it as a retriever inside your existing query engine.
Is TypeScript a constraint?: The SDK is TypeScript-first and ESM, running on Node.js 22+, Deno 2, Bun, Cloudflare Workers and browsers. LlamaIndex is strongest in Python. For a TypeScript product, the SDK avoids a second runtime in your stack.

Cortex vs LangChain

Framework or hosted retrieval.

Do you need a vector database?

What an index costs you.

Build vs framework vs SDK

The full cost picture.

SDK overview

Setup and API reference.