--- title: "Vectorize.io" type: "AI Tool" url: "https://aidemos.com/tools/vectorize-io" description: "We tested Vectorize.io on mixed docs and scans: ingestion was clean, but cited hallucinations, a Hindi miss, and freshness failures showed limits." category: "developer-tools" published: "2026-08-21T08:44:50.924166+00:00" updated: "2026-08-21T08:48:11.332938+00:00" --- # Vectorize.io Managed RAG API with clean ingestion, OCR, and public pricing, but with caveats on refusals and freshness. ## TL;DR Verdict **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:** Free $0/mo · Starter $99/mo · Pro $399/mo `Built-in OCR` · `Public pricing` · `Gmail connector` · `Reranking + filters` > **Strong managed RAG plumbing, but not a safe hallucination-free default.** > > 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. ## Demo Recording [Video: Vectorize.io demo recording](https://cdn.futuresmart.ai/public/aidemos/68a0aa71604d4ccd90ae122f87a7559e.mp4?v=1) *Video — Screen-recorded walkthrough of pipeline setup, ingestion, connector setup, live query controls, pricing, and the free-tier pipeline limit wall.* ## Feature-by-Feature Breakdown ### Mixed-Format Document Ingestion **Verdict:** Strong ingest coverage across the fixed corpus. 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. **Input:** > **File** **Output:** **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 **Verdict:** Real OCR worked, though not perfectly on every question. 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. **Input:** > **File** **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:** **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 **Verdict:** Good control surface and observability, but edge-case answer quality still failed in important places. 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:** **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 **Verdict:** Connectors worked end-to-end in the tested cases. 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:** **Input:** ``` INPUT: Walk through Notion connector setup and authorization. ``` **Output:** **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 **Verdict:** Freshness updates missed 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:** **Input:** ``` INPUT: Run the per-document Reprocess check after the overwrite. ``` **Output:** **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. | Plan | Price | Notes | | --- | --- | --- | | 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.* ## Is It Right For You? **Use it 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 it 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. ## Classification - **Category:** developer-tools - **Subcategory:** apis - **Type:** text - **Built for:** Other ## Frequently Asked Questions **Q: Does Vectorize.io OCR scanned PDFs?** 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. **Q: Does Vectorize.io refuse out-of-corpus questions reliably?** Not perfectly. It correctly refused 8 out of 10 out-of-corpus questions, but 2 questions produced reproducible cited hallucinations across all 3 rounds. **Q: Can it ingest mixed file types?** Yes. The report says md, html, csv, docx, and a Hindi PDF all ingested natively with zero extension-rejection issues. **Q: Does it support connectors like Gmail and Notion?** 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. **Q: How did document freshness behave after an overwrite?** 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. **Q: What does Vectorize.io pricing look like?** 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. **Q: Could you test multi-tenancy isolation on the free plan?** No. The free tier hit a hard cap at 1 RAG pipeline, so a second isolated tenant could not be created in this run. ## 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](mailto:contact@futuresmart.ai). ### 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](mailto:collaborate@aidemos.com).