Vectorize.io icon
developer-tools

Vectorize.io

Managed RAG API with clean ingestion, OCR, and public pricing, but with caveats on refusals and freshness.

Built-in OCRPublic pricingGmail connectorReranking + filters
TL;DR — our verdictUpdated August 2026 · 7 test artifacts

Strong managed RAG plumbing, but not a safe hallucination-free default.

Where it wins
  • You want a managed RAG pipeline that can ingest mixed document formats, including scanned PDFs, with built-in OCR.
  • You want a public, self-serve API with reranking, metadata filters, and query rewriting.
  • You want native Gmail and Notion connector support and can tolerate the free-tier pipeline limit.
Main limitation
  • You need reliably correct refusal on out-of-corpus questions with no cited hallucinations.
Pricing (verified plans)
Free $0/moStarter $99/moPro $399/mo
Strongest test artifacts
Mixed-Format Document Ingestion →Scanned PDF OCR Extraction →Retrieval API and Query Optimization →

Our take

Vectorize.io is a strong managed RAG API if you want clean ingestion across mixed document types, built-in OCR for scans, native connectors, and clear self-serve pricing. In this test it was the cleanest ingester in the session, but reproducible cited hallucinations on out-of-corpus prompts, a Hindi retrieval miss, and a freshness update failure mean you should validate it carefully before relying on it for high-stakes grounded answers.

Download video: Demo video
Screen-recorded walkthrough of pipeline setup, ingestion, connector setup, live query controls, pricing, and the free-tier pipeline limit wall.

In-Depth Review

Our detailed analysis of Vectorize.io — features, performance, and real-world testing.

AD
AI Demos Team
Expert Reviewer
Verified Review

Feature-by-Feature Breakdown

Mixed-Format Document Ingestion
Strong ingest coverage across the fixed corpus.
▾
Test Summary
Feature tested: Mixed-Format Document Ingestion
Result: Passed — Strong ingest coverage across the fixed corpus.

Feature tested: Mixed-Format Document Ingestion

Result: Passed

Verdict: Strong ingest coverage across the fixed corpus.

Expected behavior: Ingests uploaded documents across mixed file types into a RAG pipeline, including the table-heavy PDF and the md/html/csv/docx/Hindi corpus set used in the report. It also exposes visibility into document counts and processing state during the initial ingest.

Test case: PDF document → Image

Input type: PDF document

Input used: Input artifact (PDF document): INPUT: Table-heavy PDF (Sumitomo Heavy Industries Q1-2025) in the fixed corpus. — 01_tables_sumitomo-heavy-industries-Q1-2025.pdf

Observed output: Output artifact (Image): Documents tab shows the corpus ingested with 12 documents, 1,103 vectors, and 358 pages processed in sync. — 29_documents_tab.png

Input artifact: Input artifact (PDF document): INPUT: Table-heavy PDF (Sumitomo Heavy Industries Q1-2025) in the fixed corpus. — 01_tables_sumitomo-heavy-industries-Q1-2025.pdf

Output artifact: Output artifact (Image): Documents tab shows the corpus ingested with 12 documents, 1,103 vectors, and 358 pages processed in sync. — 29_documents_tab.png

What changed: PDF document transformed into Image

Test case: Text prompt → Text prompt

Input type: Text prompt

Input used: Input artifact (Text prompt): Input

Observed output: Output artifact (Text prompt): Output

Input artifact: Input artifact (Text prompt): Input

Output artifact: Output artifact (Text prompt): Output

What changed: Text prompt transformed into Text prompt

Why it matters / Conclusion: Clean first-attempt ingestion across the corpus, with strong visibility into document counts and processing state.

Ingests uploaded documents across mixed file types into a RAG pipeline, including the table-heavy PDF and the md/html/csv/docx/Hindi corpus set used in the report. It also exposes visibility into document counts and processing state during the initial ingest.

file
01_tables_sumitomo-heavy-industries-Q1-2025.pdf
INPUT: Table-heavy PDF (Sumitomo Heavy Industries Q1-2025) in the fixed corpus.
↓
OUTPUT
Output artifact for "Mixed-Format Document Ingestion" test: Documents tab shows the corpus ingested with 12 documents, 1,103 vectors, and 358 pages processed in sync., 29_documents_tab.png
Documents tab shows the corpus ingested with 12 documents, 1,103 vectors, and 358 pages processed in sync.
INPUT
INPUT: Mixed formats corpus (Markdown, HTML, CSV, DOCX, and a Hindi PDF).
↓
OUTPUT
All of those formats ingested natively with zero extension-rejection issues.
Bottom Line
Clean first-attempt ingestion across the corpus, with strong visibility into document counts and processing state.
Scanned PDF OCR Extraction
Real OCR worked, though not perfectly on every question.
▾
Test Summary
Feature tested: Scanned PDF OCR Extraction
Result: Passed — Real OCR worked, though not perfectly on every question.

Feature tested: Scanned PDF OCR Extraction

Result: Passed

Verdict: Real OCR worked, though not perfectly on every question.

Expected behavior: Reads scanned or image-only PDFs using a mixed extraction strategy with Iris vision OCR, without requiring manual preprocessing. It was exercised on the scanned test set and improved results there.

Test case: PDF document → Text prompt

Input type: PDF document

Input used: Input artifact (PDF document): INPUT: Scanned/image-only PDF from the fixed corpus. — 02_scanned_mountain-pine-beetle_USDA-1983.pdf

Observed output: Output artifact (Text prompt): Output

Input artifact: Input artifact (PDF document): INPUT: Scanned/image-only PDF from the fixed corpus. — 02_scanned_mountain-pine-beetle_USDA-1983.pdf

Output artifact: Output artifact (Text prompt): Output

What changed: PDF document transformed into Text prompt

Test case: Text prompt → Image

Input type: Text prompt

Input used: Input artifact (Text prompt): Input

Observed output: Output artifact (Image): The extraction strategy picker exposes Fast, Iris, and Mixed, with Mixed selected for this pipeline. — 08_extraction_strategy_options.png

Input artifact: Input artifact (Text prompt): Input

Output artifact: Output artifact (Image): The extraction strategy picker exposes Fast, Iris, and Mixed, with Mixed selected for this pipeline. — 08_extraction_strategy_options.png

What changed: Text prompt transformed into Image

Why it matters / Conclusion: This is a genuine scanned-PDF OCR path, and it materially improved results on the scanned test set.

Reads scanned or image-only PDFs using a mixed extraction strategy with Iris vision OCR, without requiring manual preprocessing. It was exercised on the scanned test set and improved results there.

file
02_scanned_mountain-pine-beetle_USDA-1983.pdf
INPUT: Scanned/image-only PDF from the fixed corpus.
↓
OUTPUT
Mixed extraction with Iris OCR answered 4/5 scanned-document questions correctly; only the byline question was missed.
INPUT
INPUT: Select the Mixed extraction strategy (Fast + Iris vision OCR) for the pipeline.
↓
OUTPUT
Output artifact for "Scanned PDF OCR Extraction" test: The extraction strategy picker exposes Fast, Iris, and Mixed, with Mixed selected for this pipeline., 08_extraction_strategy_options.png
The extraction strategy picker exposes Fast, Iris, and Mixed, with Mixed selected for this pipeline.
Bottom Line
This is a genuine scanned-PDF OCR path, and it materially improved results on the scanned test set.
Retrieval API and Query Optimization
Good control surface and observability, but edge-case answer quality still failed in important places.
▾
Test Summary
Feature tested: Retrieval API and Query Optimization
Result: Partial — Good control surface and observability, but edge-case answer quality still failed in important places.

Feature tested: Retrieval API and Query Optimization

Result: Partial

Verdict: Good control surface and observability, but edge-case answer quality still failed in important places.

Expected behavior: Provides a production retrieval endpoint plus chat/completions access, with reranking, metadata filters, and automatic query rewriting for grounded answers. The same endpoint was exercised in the evaluation that surfaced cited hallucinations and a missed Hindi document.

Test case: Text prompt → Text prompt

Input type: Text prompt

Input used: Input artifact (Text prompt): Input

Observed output: Output artifact (Text prompt): Output

Input artifact: Input artifact (Text prompt): Input

Output artifact: Output artifact (Text prompt): Output

What changed: Text prompt transformed into Text prompt

Test case: Text prompt → Image

Input type: Text prompt

Input used: Input artifact (Text prompt): Input

Observed output: Output artifact (Image): The Connect tab exposes the retrieval endpoint, org and pipeline IDs, token generation, metadata filtering, reranking, and query rewriting controls. — 18_connect_tab.png

Input artifact: Input artifact (Text prompt): Input

Output artifact: Output artifact (Image): The Connect tab exposes the retrieval endpoint, org and pipeline IDs, token generation, metadata filtering, reranking, and query rewriting controls. — 18_connect_tab.png

What changed: Text prompt transformed into Image

Why it matters / Conclusion: The API surface is strong, but the same endpoint still produced reproducible cited hallucinations and missed the Hindi document in this evaluation.

Provides a production retrieval endpoint plus chat/completions access, with reranking, metadata filters, and automatic query rewriting for grounded answers. The same endpoint was exercised in the evaluation that surfaced cited hallucinations and a missed Hindi document.

INPUT
INPUT: 48 labelled questions, run 3 times each, through combined /retrieval and /chat/completions API calls.
↓
OUTPUT
All 144 queries completed with 0 API failures, and the retrieval API returned full chunk text plus similarity and reranked relevancy scores, source file, and chunk/page indices.
INPUT
INPUT: Configure retrieval controls for a production pipeline.
↓
OUTPUT
Output artifact for "Retrieval API and Query Optimization" test: The Connect tab exposes the retrieval endpoint, org and pipeline IDs, token generation, metadata filtering, reranking, and query rewriting controls., 18_connect_tab.png
The Connect tab exposes the retrieval endpoint, org and pipeline IDs, token generation, metadata filtering, reranking, and query rewriting controls.
Bottom Line
The API surface is strong, but the same endpoint still produced reproducible cited hallucinations and missed the Hindi document in this evaluation.
External Source Connector Sync
Connectors worked end-to-end in the tested cases.
▾
Test Summary
Feature tested: External Source Connector Sync
Result: Passed — Connectors worked end-to-end in the tested cases.

Feature tested: External Source Connector Sync

Result: Passed

Verdict: Connectors worked end-to-end in the tested cases.

Expected behavior: Connects external sources through OAuth/connectors and syncs them into the pipeline canvas, including Gmail and Notion in the tested setup. Gmail worked end-to-end, and Notion reached consent successfully before the attempt was stopped for an account-choice reason.

Test case: Text prompt → Image

Input type: Text prompt

Input used: Input artifact (Text prompt): Input

Observed output: Output artifact (Image): The pipeline canvas now includes Gmail test as a second source, and the report says Gmail sync completed with 3 total documents and 85 new chunks. — 48_gmail_saved.png

Input artifact: Input artifact (Text prompt): Input

Output artifact: Output artifact (Image): The pipeline canvas now includes Gmail test as a second source, and the report says Gmail sync completed with 3 total documents and 85 new chunks. — 48_gmail_saved.png

What changed: Text prompt transformed into Image

Test case: Text prompt → Image

Input type: Text prompt

Input used: Input artifact (Text prompt): Input

Observed output: Output artifact (Image): The Notion integration modal reaches the Authorize step with a filled connector name and no platform errors. — 39_notion_authorize.png

Input artifact: Input artifact (Text prompt): Input

Output artifact: Output artifact (Image): The Notion integration modal reaches the Authorize step with a filled connector name and no platform errors. — 39_notion_authorize.png

What changed: Text prompt transformed into Image

Why it matters / Conclusion: Gmail worked end-to-end, and Notion reached consent cleanly before the attempt was abandoned for an account-choice reason rather than a platform failure.

Connects external sources through OAuth/connectors and syncs them into the pipeline canvas, including Gmail and Notion in the tested setup. Gmail worked end-to-end, and Notion reached consent successfully before the attempt was stopped for an account-choice reason.

INPUT
INPUT: Add a live Gmail OAuth connector to the existing File Upload pipeline.
↓
OUTPUT
Output artifact for "External Source Connector Sync" test: The pipeline canvas now includes Gmail test as a second source, and the report says Gmail sync completed with 3 total documents and 85 new chunks., 48_gmail_saved.png
The pipeline canvas now includes Gmail test as a second source, and the report says Gmail sync completed with 3 total documents and 85 new chunks.
INPUT
INPUT: Walk through Notion connector setup and authorization.
↓
OUTPUT
Output artifact for "External Source Connector Sync" test: The Notion integration modal reaches the Authorize step with a filled connector name and no platform errors., 39_notion_authorize.png
The Notion integration modal reaches the Authorize step with a filled connector name and no platform errors.
Bottom Line
Gmail worked end-to-end, and Notion reached consent cleanly before the attempt was abandoned for an account-choice reason rather than a platform failure.
Incremental Sync and Document Reprocessing
Freshness updates missed the changed content within the observed window.
▾
Test Summary
Feature tested: Incremental Sync and Document Reprocessing
Result: Failed — Freshness updates missed the changed content within the observed window.

Feature tested: Incremental Sync and Document Reprocessing

Result: Failed

Verdict: Freshness updates missed the changed content within the observed window.

Expected behavior: Supports explicit update checks and per-document reprocessing to refresh previously ingested content. In the observed window, same-filename content updates did not propagate.

Test case: Text prompt → Image

Input type: Text prompt

Input used: Input artifact (Text prompt): Input

Observed output: Output artifact (Image): After sync, the pipeline stayed Listening and the log reported 'no changes' even though the file content had changed. — 28_after_sync_trigger.png

Input artifact: Input artifact (Text prompt): Input

Output artifact: Output artifact (Image): After sync, the pipeline stayed Listening and the log reported 'no changes' even though the file content had changed. — 28_after_sync_trigger.png

What changed: Text prompt transformed into Image

Test case: Text prompt → Image

Input type: Text prompt

Input used: Input artifact (Text prompt): Input

Observed output: Output artifact (Image): The pipeline is Idle and scheduled reprocessing completed, but the canary string was still never retrievable after more than 5 minutes. — 31_reprocess_check1.png

Input artifact: Input artifact (Text prompt): Input

Output artifact: Output artifact (Image): The pipeline is Idle and scheduled reprocessing completed, but the canary string was still never retrievable after more than 5 minutes. — 31_reprocess_check1.png

What changed: Text prompt transformed into Image

Why it matters / Conclusion: Confirmed freshness weakness: neither update mechanism propagated the changed content within the observed window.

Supports explicit update checks and per-document reprocessing to refresh previously ingested content. In the observed window, same-filename content updates did not propagate.

INPUT
INPUT: Overwrite an indexed document with a same-filename canary update, then trigger Check for updates.
↓
OUTPUT
Output artifact for "Incremental Sync and Document Reprocessing" test: After sync, the pipeline stayed Listening and the log reported 'no changes' even though the file content had changed., 28_after_sync_trigger.png
After sync, the pipeline stayed Listening and the log reported 'no changes' even though the file content had changed.
INPUT
INPUT: Run the per-document Reprocess check after the overwrite.
↓
OUTPUT
Output artifact for "Incremental Sync and Document Reprocessing" test: The pipeline is Idle and scheduled reprocessing completed, but the canary string was still never retrievable after more than 5 minutes., 31_reprocess_check1.png
The pipeline is Idle and scheduled reprocessing completed, but the canary string was still never retrievable after more than 5 minutes.
Bottom Line
Confirmed freshness weakness: neither update mechanism propagated the changed content within the observed window.

Public self-serve pricing

Free, Starter, and Pro are shown directly on the page.

Free
$0/mo
Includes 1,500 pages per month and 500 queries per month.
Starter
$99/mo
Highlighted on the pricing page. Includes 15,000 pages per month and 1,000 queries per month, with overages listed at $0.02/page and $0.005/query.
Pro
$399/mo
Price shown on the page; the report did not state a usage-quota breakdown for this plan.

The page shows public pricing without a sales call. The comparison table also shows Enterprise, but no Enterprise price was stated in the report.

✓ Use This If
●You want a managed RAG pipeline that can ingest mixed document formats, including scanned PDFs, with built-in OCR.
●You want a public, self-serve API with reranking, metadata filters, and query rewriting.
●You want native Gmail and Notion connector support and can tolerate the free-tier pipeline limit.
✕ Skip This If
●You need reliably correct refusal on out-of-corpus questions with no cited hallucinations.
●Fresh same-filename document updates must appear quickly in retrieval results.
●You need proven cross-tenant isolation on the free plan or want to test multiple tenants immediately.
developer-toolsapistextOther
Yes. In Mixed mode with Iris vision OCR, the scanned/image-only PDF answered 4 out of 5 scanned-document questions correctly; only the byline question was missed.
Not perfectly. It correctly refused 8 out of 10 out-of-corpus questions, but 2 questions produced reproducible cited hallucinations across all 3 rounds.
Yes. The report says md, html, csv, docx, and a Hindi PDF all ingested natively with zero extension-rejection issues.
Yes. A live Gmail OAuth connector worked end-to-end, and the Notion flow reached the authorization step cleanly before being abandoned for an account-choice reason.
Poorly in this test. A same-filename update did not propagate through either Check for updates or per-document Reprocess within more than 5 minutes, and the canary string never became retrievable.
The pricing page showed Free at $0/month, Starter at $99/month, and Pro at $399/month. The report also noted Free includes 1,500 pages and 500 queries per month, while Starter includes 15,000 pages and 1,000 queries per month with listed overages.
No. The free tier hit a hard cap at 1 RAG pipeline, so a second isolated tenant could not be created in this run.

Banner Preview

How the embed badge will look on your site

Vectorize.io featured on AI Demos

Embed HTML

Copy this code to your website source

<a target="_blank" href="https://aidemos.com/tools/vectorize-io?utm_source=vectorize-io_embed" style="width: 250px; height: 80px; border-radius:4px;" width="250" height="80"> <img src="https://aidemos-website-images.s3.amazonaws.com/featured.png" alt="Vectorize.io | Featured on AI Demos" style="width: 250px; height: 80px; border-radius:4px;" width="250" height="80"> </a>

Quick Integration Guide

  • 1Copy the HTML code block above.
  • 2Paste it into your site's HTML or CMS editor.
  • 3Banner appears instantly on your page.
  • 4Links back to your tool profile here.
Similar Tools

Similar Tools

Discover more AI tools like Vectorize.io to enhance your workflow.

🤖
Vectara
Vectara gives you a grounded-answer API with strong refusals, but tables, contradictions, scans, and pricing are the weak spots.
AI Tool
🤖
RAGFlow Cloud
AI Tool
🤖
Dify
API-first managed RAG with strong table answers and connectors, but you still need guardrails for scans, multilingual retrieval, and refusal cases.
AI Tool
🤖
CustomGPT AI
AI Tool

Comments (0)

Please Log in to join the discussion.

Built by FutureSmart AI — the team behind AI Demos

Need a custom AI solution for this use case?

If you are looking to build a custom document ingestion, OCR, or RAG pipeline for your business or internal workflow, email us at contact@futuresmart.ai.

Get a custom build

Found something inaccurate or missing? We try to keep our AI research accurate and useful. If you found outdated information, an issue, or have a suggestion, email us at collaborate@aidemos.com.

Back to Top