Get Demo Center access

Head-to-head · updated 13 September 2026

DataShield vs Ragie: who should run the RAG pipeline, them or you?

Ragie sells the RAG pipeline you did not want to build. Their own line: "The context engine for agents, assistants and apps." Push a PDF, a Slack export or a video at the API and it comes back as clean, cited context. The parsing is the good part. Their agentic OCR pulls tables, forms and charts out of ugly scans with bounding boxes. Agentic Retrieval splits a hard question into parts, checks its own answers, then cites them. They shipped an MCP server in March 2025, before most of this market could spell the acronym.

One thing first, because you will find it anyway. Today their homepage carries a banner that reads "Ragie service will end on July 19." Their app and status subdomains no longer resolve. We found no post explaining it, so ask them before you plan anything. We take no pleasure in that, and the rest of this page is not about it. It is about the design choice every managed context engine makes: your documents go to their index, and their control plane decides what comes back. Ours runs the other way. The corpus sits on your own hardware, embeddings are computed locally, and the record of what each agent read is a hash chain you can check without trusting us.

Get free Demo Center access Get an instant quote
DataShield vs Ragie at a glanceEight questions a regulated buyer asks a RAG vendor. Scored from each side's public material. DataShield vs Ragie at a glance Eight questions a regulated buyer asks a RAG vendor. Scored from each side's public material. DataShield Ragie Tamper-evident record of what an agent retrieved PII and PHI tokenized before the embedding model Embeddings computed locally, no text egress Runs on your own boxes, no vendor control plane Document parsing, OCR and multimodal ingest Managed connector catalogue SOC 2 Type II, HIPAA, GDPR Prices published before a sales call shipped partial / roadmap not offered Sources at the bottom of this page.

The short version

Pick DataShield when

  • The documents have patient names, account numbers or case files in them. Datasets are tokenized at ingest; agents query tokenized data over MCP; detokenization is a privileged, audited operation.
  • Nothing may leave the building, including the text you send off to be turned into vectors. We embed locally with nomic-embed-text, so the chunks never cross your border. How the stack fits together.
  • Someone will one day ask you to prove the retrieval log was not edited. Our chain answers that with math. Run the verifier.
  • You want to pull an agent's authority mid-session and have the very next tool call fail closed. How Auth does it.

Pick Ragie when

  • Your inputs are messy scans, forms and tables. Their agentic OCR is better at that than ours, and it is not close.
  • You need audio and video in the same index as the text, through one pipeline, with no work from you.
  • You want SharePoint, Slack, Notion, Confluence, Drive, Zendesk and friends syncing by tomorrow, and you would rather pay $250 a month per connector than write them.
  • Your team is three people and the whole point is that nobody runs a vector database. Their free tier gets you a working app before lunch.

Bottom line: Ragie is better at turning documents into context. We are built for the case where the documents are sensitive, the buyer is regulated, and the agent has to be told no. If you are reading this because the banner on their homepage sent you looking, start at /demos and bring one of your worst PDFs.

Feature by feature: managed RAG-as-a-Service against a self-hosted corpus

Competitor cells describe what Ragie's public site, pricing page and launch posts say as of the date above. If we have mischaracterised something, email support@myorg.ai and we will correct it, credited.

What mattersDataShieldRagieEdge
Document parsing and multimodal ingestText extraction from PDF, DOCX and EML, stored as a versioned derived file. Ingest from inline uploads, a file repo, guarded URLs, or streamed straight out of S3-compatible storage. No OCR, no audio, no video.Agentic OCR with tables, forms, charts and bounding boxes, plus native audio and video retrieval through the same pipeline. This is their best work.◇
Managed connectorsNone of that. We read from object storage and from the sources you register as providers. Slack and Confluence are your problem.Ragie Connect covers Drive, Notion, Confluence, Slack, SharePoint, Dropbox, Intercom, Zendesk and more, with embeddable connectors so your customers link their own accounts.◇
Retrieval engineeringHybrid retrieval over pgvector: a keyword leg and a vector leg, 768 dimensions, with incremental reindexing. Solid, not a research programme.Three indexes (vector, keyword, summary), reranking, recency bias, deep search, and Agentic Retrieval that decomposes a query and grades its own evidence.◇
Tokenization and data handlingDatasets are tokenized at ingest; agents query tokenized data over MCP; detokenization is a privileged, audited operation. Tokens are deterministic and join-preserving, so a join still works after the names are gone. Quasi-identifier generalization covers dates to year, decade or age band, ZIPs to 3 or 4 digits, and partial phones, SSNs and emails, with a measured cardinality-reduction score per column.Partition-level isolation by tenant, workspace or customer, and AES-256 at rest. We found no field-level tokenization or masking before indexing in their public docs.◆
Embeddings and egressnomic-embed-text runs locally. Chunk text never leaves your network to be vectorized. Every dataset value bound for a prompt also crosses a fail-closed gate: a PII or PHI column with no configured treatment is redacted, and a PHI dataset refuses an AI endpoint that is not marked BAA or ZDR approved.Their managed pipeline does the parsing, chunking and embedding, which means the document text lands in their service first. That is the deal you sign.◆
Audit evidenceSHA-256 hash chain with Ed25519-signed checkpoints that are themselves chained. Verification names the failure: tampering, insertion, deletion, or truncation. Try the verifier.Not described in their public material. Ask what record survives of which agent retrieved which chunk, and who can edit it.◆
Agent authorizationEvery governed tool call passes a scope ceiling, a consented-tool allowlist, an authority tier and a revocation re-check before dispatch. It fails closed, and the call is metered and attributed to the agent.Partition scoping decides which slice of the index a key can see. That is coarse access control, set once. We found no per-call authority check.◆
Break-glassScoped, time-boxed emergency access for agents. It auto-revokes and cannot be quietly deleted from the log.Not described in their public material.◆
GDPR erasureCrypto-shred of per-subject key material plus ISO 27560 consent receipts. The audit chain still verifies after the subject is gone.Documents can be deleted and they state they do not train on your data. What happens to derived chunks, embeddings and summaries is not spelled out. Ask.◆
MCP and agentsMore than 200 MCP tools across Ontology, Auth, Corpus and Lighthouse. Auth issues MCP tool tokens with scope ceilings, and delegation is RFC 8693 token exchange with an enforced ceiling.A hosted Streamable HTTP MCP server with context-aware tool descriptions that change as the knowledge base changes, so an agent picks retrieval over web search. Genuinely thoughtful, and shipped early.◈
DeploymentSelf-hosted in your own cloud or data center, or a dedicated single-tenant server we operate. Docker images for Auth, Ontology, Corpus and Lighthouse, with a signed deploy manifest Guardian verifies.Their site offers cloud, your own VPC, or on-prem. We could not test the last two, and the developer path is clearly the hosted one.◈
CertificationsSOC 2 not yet certified, and we will not imply otherwise. Our standards work is design mapping: HIPAA §164.312(b), §164.514, GDPR Art. 17, ISO 27560, RFC 8785.SOC 2 Type II, GDPR, HIPAA and CCPA, stated on the homepage and in their October 2025 launch post. They beat us here today.◇
PricingPublished model and a scoped instant quote, no sales wall.Published too: free developer tier, $100 a month Starter, $500 a month Pro, quote-only Enterprise, with per-page, storage and connector rates listed.◈

◆ DataShield leads◇ Ragie leads◈ comparable

Ragie claims are drawn from ragie.ai, its pricing page and its own launch posts, last checked 13 September 2026. We link them below rather than work from memory.

Three things you get here that you won't get from a managed RAG API

The model sees tokens, not names

Most RAG services index what you give them. If the contract has a member ID in paragraph four, the member ID is in the chunk, in the embedding, and one day in a prompt. We tokenize at ingest, keep the tokens join-preserving so analysis still works, and make detokenization a privileged call that gets logged. What the catalog does first.

Proof an examiner can check

A retrieval log that can be edited proves nothing. Ours is a hash chain with signed checkpoints, and the verifier says what broke: tampering, insertion, deletion or truncation. That is the property EU AI Act Article 12 and HIPAA §164.312(b) reviewers keep asking about. Try it in your browser, no signup.

Authority that changes mid-flight

An analyst resigns at 4pm on a Friday. Their agent is 20 minutes into a 40-minute research job over your case files. With DataShield the next governed call is re-checked against current authority and fails closed. An API key scoped to a partition keeps working until someone rotates it. How Auth does it.

Where Ragie is genuinely stronger

We should say this plainly. Ragie is better at the thing most people mean by RAG. Their parser reads a scanned invoice and gives you the table, with coordinates. Audio and video go in the same door as the PDFs. Agentic Retrieval takes a question like "how did gross margin move between 2021 and 2023, and how does that compare to rivals", splits it, gathers evidence, throws out the false positives and cites the rest. They open-sourced a reference chatbot, they open-sourced their context-aware MCP tooling, and in April 2026 they shipped a Claude Code skill so an agent could write Ragie code by itself. Developers liked them for good reasons, and the testimonial wall reads like real people rather than case-study prose. We do not match their parsing and we are not pretending to.

Here is the push-back, and it has nothing to do with the banner. A managed context engine is a second copy of your documents somewhere else, with an access model that is set at key-issue time and a log you cannot independently verify. That is a fine trade when the corpus is product docs. It is a different trade when the corpus is clinical notes or a litigation hold. Partition isolation answers "which tenant", not "was this agent allowed to read this chunk at 14:07 on 3 March, and can you prove nobody edited the record since". Those are the questions that turn up eighteen months later, usually in writing, usually from someone with a deadline.

Questions worth asking both of us

These are the questions we would want answered if we were the ones buying. Ask them on every call, ours included.

Is Ragie actually shutting down?

We do not know, and we will not claim more than we can show. On 13 September 2026 their homepage carried a banner reading "Ragie service will end on July 19." Their app and status subdomains did not resolve, their docs host did not answer, their last changelog entry was 5 June 2026 and their last blog post was 14 April 2026. There was no announcement post we could find. Email support@ragie.ai and get it in writing before you plan a migration or a renewal.

Can you cryptographically prove a retrieval log entry wasn't deleted?

DataShield: yes. Each record commits to the one before it, checkpoints are signed and chained, and verification tells deletion apart from truncation and from tampering. Run it against a sample chain at /verify. Ragie: their public material describes encryption and partition isolation, not tamper evidence. Ask any RAG vendor to show you the mechanism, not the compliance badge.

What happens to a revoked agent mid-session?

DataShield re-checks authority on every governed tool call, so revocation lands on the next call rather than the next token refresh. With a hosted RAG API the usual answer is an API key scoped to a partition, which keeps working until a human rotates it. Ask how long a compromised key keeps reading your index.

How does GDPR erasure interact with the audit trail?

DataShield crypto-shreds per-subject key material and issues an ISO 27560 consent receipt. Actor identities in the chain are HMAC-committed, so the evidence still verifies once the subject is gone. Ragie lets you delete documents and says it does not train on your data. What happens to chunks, summaries and embeddings derived from a deleted document is not described. That derived layer is where erasure requests usually go wrong.

We're on Ragie now. What does moving to DataShield actually involve?

Your files, not their index. You re-ingest from wherever the documents really live: object storage, a URL list, or direct upload. Ontology classifies the fields against 129 PII and PHI classes, you decide what gets tokenized or generalized, then the corpus is built with local embeddings and hybrid retrieval. Agents reach it over MCP under Auth tool tokens. Expect to lose OCR quality on scanned documents and to gain a corpus that runs on your hardware. Bring your three worst files to the Demo Center and we will tell you honestly whether the parsing holds up.

Does DataShield have SOC 2?

Not yet, and we say so on every page. Ragie has SOC 2 Type II and we do not. What we have instead is live in production (Guardian and Lighthouse since April 2026), a public security page with a threat model, a verifier anyone can run, and design-partner terms that include source escrow, so a small vendor is not a single point of failure. Weigh that how you like, but weigh it with the badge question the right way round: a badge says a process was audited, not that your agent was told no.

Other head-to-heads

RAG platform

DataShield vs Vectara

A managed RAG stack, and the governance it leaves to you.

 Framework

DataShield vs LlamaIndex

Build your own pipeline, or buy one that is already governed.

 Vector DB

DataShield vs Pinecone

Vector search at scale, versus tokenized data and proof.

 All

Every comparison

One honest scorecard per vendor.

See both mechanisms run in your browser: break a live audit chain, revoke an agent mid-session, then decide where your documents should actually live. Demo Center access is free with a work email.

Get free Demo Center access

You've seen the proof

Ready for a number? Scope your deployment and we'll price it against your own economics.

Get your quote →