Docubix
Docubix is a hosted retrieval-augmented generation service: upload PDF, DOCX, TXT or Markdown files, and a single REST endpoint returns answers cited back to the exact document and page. A free tier is permanent; Pro costs 79 USD monthly.
What is Docubix?
Docubix is a hosted retrieval-augmented generation service aimed squarely at people who would otherwise assemble a RAG stack themselves. The product is organised around three steps. You upload documents in PDF, DOCX, TXT or Markdown, and the platform parses, chunks, embeds and indexes them without further instruction. You then configure the assistant: a system prompt, a model, and retrieval settings such as top-k and score thresholds. Finally you integrate, calling a REST endpoint from your own application or backend.
The unit of work is the knowledge base. Each one carries its own vector index, its own settings and its own API key, so several projects sit side by side without leaking into one another. The API surface covers knowledge bases, document upload, API keys, conversations, search and chat. Two authentication schemes coexist: dashboard operations send a Clerk user header, while public chat carries the API key in the JSON body.
What distinguishes the responses is that citations are mandatory. Every answer comes back with a sources array naming the document, the chunk, its similarity score and the page or section it came from, alongside token usage and retrieval metadata. Retrieval runs on cosine similarity with score thresholds, and the vendor says a fallback response is returned when no chunk clears the bar. Answers can stream token by token or arrive as a complete payload. A web dashboard handles bases, documents, keys and conversation review, and reports on which questions are asked and which documents get cited most.
The site names four audiences: SaaS products and apps, customer support teams, internal wikis and onboarding, and course platforms. Embedding and model configuration stay server-side and are not exposed. The product is young and says so, describing itself in the footer as made for developers, in beta, with a domain registered in June 2026 and a blog that is still empty.
What it does
- Turn a folder of documents into a queryable assistant without building a retrieval pipeline
- Answer questions from your own content and cite the exact document and page behind each answer
- Serve those answers inside your product through a single REST call
- Stream responses token by token instead of waiting for a complete payload
- Isolate projects behind separate knowledge bases, vector indexes and API keys
- Keep conversation history so users can pick up where they left off
- See which questions are asked, which documents get cited and where the documentation falls short
When to use Docubix / When not to
A quick filter to help you decide if Docubix is the right fit.
When to use Docubix
- Startup founders adding a documentation-aware assistant to a SaaS product without hiring an ML engineer
- Backend and full-stack developers who want retrieval, generation and citations behind a single REST call
- Customer support teams deflecting repetitive tickets with answers that cite the help-centre article used
- Internal operations and HR teams turning handbooks, wikis and SOPs into something staff can simply ask
- Technical writers and documentation managers replacing docs search with a question-answering layer
When not to use Docubix
- Organisations bound by data-residency rules, since no hosting country or region is named anywhere on the site
- Buyers who need a signed data processing agreement, an explicit GDPR commitment or a security certification before procurement will sign off
- Teams handling health, financial or otherwise regulated records, which the terms and the privacy policy explicitly exclude
- Non-technical buyers expecting a paste-in widget, because every integration goes through server-side REST calls
- Products with sustained query volume, as the top published plan stops at 3,000 queries a month and 1,000 documents
How to use Docubix
A typical end-to-end flow, from setup to results.
- Create an account on the Docubix dashboard and sign in
- Create a knowledge base with a POST to the /kbs endpoint, sending the Clerk user header and a name
- Upload a PDF, DOCX, TXT or Markdown file to that base through the documents endpoint
- Poll the document status until it reads completed, because processing runs asynchronously
- Set the system prompt, the model and the retrieval settings for that knowledge base
- Generate an API key for the base and store it immediately, as it is displayed only once
- Ask a first question by posting the API key and the message to the chat endpoint
- Read the answer together with its sources array, and surface those citations in your own interface
- Continue the thread by sending back the conversation identifier returned by the previous response
- Move to the streaming endpoint once the flow works, keeping every call server-side so the key is never exposed
Pros & Cons
Pros
- Citations are not an add-on: every answer names its source document and page, on the free plan as well as on Pro
- The free tier is permanent rather than a countdown trial, includes API access and needs no payment card
- Pricing is two flat monthly plans with no token arithmetic to model
- No vector database to run and no ML expertise required, since embedding and retrieval are managed
- The API is properly documented: real endpoints, a full response schema, a dedicated errors page and samples in cURL, JavaScript, Next.js and Python
- Retrieval metadata is exposed in the response, which makes answers debuggable and auditable
- Knowledge bases are isolated, with scoped keys and per-base rate limits
Cons
- The product is openly in beta, the domain was registered in June 2026, and there is no track record to check
- No legal entity, postal address or phone number is published, and a single support mailbox is the only channel
- GDPR is never mentioned, no data processing agreement is offered and no certification is claimed
- No hosting country or region is disclosed, so data residency cannot be established
- The terms disclaim any availability guarantee and cap liability at 100 USD where no fees have been paid
- Only PDF, DOCX, TXT and Markdown are ingested; HTML and a Notion sync are announced but unavailable
- The two published plans stop at 3,000 queries a month, and the site contradicts itself on how many knowledge bases Pro allows
Pricing & Plans
There is a permanent free plan rather than a time-limited trial, and it requires no payment card. It covers one knowledge base, twenty documents, one hundred queries per month and 0.25 GB of storage, with API access and citations included. The cheapest paid entry point is the Pro plan at 79.00 USD per month, billed monthly; no annual rate, enterprise tier or usage-based option is published. Payments are processed by Paddle, refunds follow Paddle's policies and applicable consumer protection law, and approved refunds return to the original payment method. A subscription can be cancelled at any time and stays active until the end of the current billing period.
- Free — 0 USD per month — 1 knowledge base
- up to 20 documents
- 100 queries per month
- 0.25 GB storage
- API access
- citations on every answer
- Pro — 79 USD per month — 10 knowledge bases
- up to 1
- 000 documents
- 3
- 000 queries per month
- 10 GB storage
- API access
- citations on every answer
Data, GDPR & hosting
A consolidated view of how Docubix handles your data.
GDPR overview
There is no GDPR implementation to describe, and that has to be said plainly: the word GDPR appears nowhere on the site, and neither do CCPA, SOC 2, ISO 27001 or HIPAA. What exists is a rights paragraph in the privacy policy covering access, correction, deletion, objection and a copy of one's information, introduced by the qualifier depending on your jurisdiction and exercised by writing to the general support address. International transfers are acknowledged in one sentence, with no transfer mechanism or standard contractual clauses named. No Article 27 representative is designated, no data protection officer is named, no postal address is published and no data processing agreement is offered. A European buyer should treat compliance here as undocumented rather than established.
Who owns the data?
The terms are unambiguous on ownership. Users keep all rights to the documents, data and prompts they upload, and nothing in the agreement transfers that content to Docubix. What the customer grants instead is a limited, non-exclusive, worldwide licence letting the vendor store, process, index, embed, retrieve and transmit that content, and the terms state this licence exists solely to provide and improve the service. The platform itself, including its software, design, branding and underlying technology, remains the property of Docubix or its licensors. Users are separately responsible for holding the rights to whatever they upload.
Reuse rights
Reuse needs no permission. Because ownership of uploaded documents, prompts and data never passes to the vendor, customers keep the right to use their own content and the answers built from it as they see fit, and nothing in the terms conditions that reuse. The obligations run the other way. Users must hold the rights to whatever they upload, must keep health, financial and other regulated records off the platform, and remain responsible for verifying AI outputs, which the terms warn may be inaccurate. The restrictions bite on the platform rather than on the output: the service may not be copied, modified, reverse-engineered or turned into derivative works, its proprietary technology may not be extracted, and rate limits may not be circumvented. On the vendor's side, submitted content may be processed by third-party AI providers the site does not name, but Docubix states it does not train its own models on customer content and does not sell personal information.
Data retention & training
Hosting summary
Docubix does not say where the data lives. No page names a hosting country, a region or a cloud provider; the privacy policy lists cloud hosting and data storage only as categories of service provider, without identifying either. The single geographic statement acknowledges that information may be processed and stored in countries other than the user's country of residence, and asks the user to accept that such transfers may occur. Other vendors are named in the same section, including Clerk for authentication, Paddle for payments, Google Analytics, and PostHog for in-app product analytics and session replay, but hosting is not among them. There is no trust page, no security page and no status page. For what it is worth, the domain resolves to an anycast node on Amazon's network located in the United States, but that is an infrastructure observation rather than a residency commitment, and it says nothing about where documents and embeddings are actually stored. Anyone with a residency requirement should ask before uploading.
Things to keep in mind
Risks and trade-offs to weigh before adopting Docubix.
- Cited answers feel authoritative, yet the terms warn that outputs may be wrong and must be reviewed; an agent who stops opening the source will eventually pass on a confident error
- A citation shows where an answer came from, not whether the document is still correct, so stale documentation produces stale answers with a footnote attached
- Uploading a handbook or a help centre hands that corpus to third-party AI providers the site does not name
- With no hosting country, no data processing agreement and no GDPR wording, a European deployment rests entirely on the buyer's own risk assessment
- The API key travels in the request body, so a single call made from browser code exposes it to every visitor
- Liability is capped at 100 USD where no fees have been paid and availability is not guaranteed, which leaves an assistant embedded in a product with no contractual recourse if it goes down
- Session recording runs on signed-in users of the application; masking of sensitive areas is declared but cannot be verified from outside
Setup & Integrations
Technical difficulty
Moderate, and the promise and the reality differ. Nothing needs installing, no vector database has to be run and no ML knowledge is required; the quick start claims a first answer in about five minutes. But every integration is server-side code: REST calls, key handling, and an explicit warning never to call the API from a browser. There is no official SDK, only copy-paste samples in cURL, JavaScript, Next.js and Python. Document processing is asynchronous, so status must be polled, and two authentication schemes coexist. A developer will be comfortable here; a non-technical buyer will not.
Deployment
Integrations
Behind Docubix
Resources
All the official URLs gathered for verification and reference.
Alternatives
Tools that compete with or complement Docubix.
Frequently asked questions
Which file formats can I upload?
Do answers always come with citations?
Can I call the API directly?
How many knowledge bases do I get?
Do I have to train or fine-tune a model?
Which embedding model does Docubix use?
Is my content used to train AI models?
What does it cost, and can I cancel?
Is there a minimum age?
Is Docubix ready for production use?
Should you pick Docubix?
Docubix does one thing and describes it honestly: it removes the work of assembling a retrieval pipeline. The engineering on show is credible, with real endpoints, a complete response schema, retrieval metadata exposed for debugging, streaming alongside standard calls, and code samples in four environments. Making citations compulsory on every answer, including on the free tier, is the decision that most separates it from a generic chatbot builder, and it is what makes the tool defensible in a support or documentation setting where an answer has to be traceable.
The reservations are not about the product but about the company behind it. No legal entity is named, no address is published, and the only way to reach anyone is a single support mailbox. The terms place governing law in India and jurisdiction in Hyderabad, yet nothing on the site establishes where the business actually sits. GDPR is never mentioned, no data processing agreement is offered, and no hosting country or region is disclosed. Availability is not guaranteed and liability is capped at 100 USD where no fees have been paid. The domain was registered in June 2026, the footer says beta, and the blog is still empty.
Read together, that gives a fairly clear verdict. For a founder wiring cited answers into an MVP, an internal wiki assistant, or a support bot working on non-sensitive content, the free tier costs nothing to evaluate and the 79 USD plan is legible. For anyone with a procurement checklist, a residency requirement or regulated data, the gaps are not details to be settled later: they are unpublished today, and would have to be resolved directly with the vendor before anything is uploaded.
- Choosing a selection results in a full page refresh.
- Opens in a new window.