Guide

White-label RAG for agencies and ESNs: licence, BYOK, deployment at the client

An agency or ESN (IT services company) that delivers a RAG under its own brand makes four decisions before the first client. Who is responsible for which data? How are clients isolated from each other? Who pays for the model? Where does the platform run? With IgnitionRAG, white-labelling is reserved for the Agency/ESN licence, quoted on request, with BYOK available, deployable at the client or with IgnitionAI.

Transparency note: this guide is published by IgnitionAI, the company behind IgnitionRAG. GDPR references rely on the guidelines of the European Data Protection Board. Security references link to OWASP publications. All were consulted on 22 October 2026. This guide contains no licence price and no numerical estimate.

What "white label" covers in IgnitionRAG, factually

In IgnitionRAG, white-labelling lets you change the colour, the name and the logo from the admin console. Your clients see your brand.

It is sold with the Agency/ESN licence. It is not part of the Solo (€59 excl. VAT/month) and Team (€249 excl. VAT/month) SaaS plans. The licence price is quoted on request.

ItemSaaS Solo / TeamAgency/ESN licence
Editable colour, name, logoNoYes
Model keysManaged AI, credits included, no keyBYOK available (your keys or the client's)
Deployment locationIgnitionRAG infrastructure, OVH FranceClient cloud (Azure, AWS, GCP) or IgnitionAI infrastructure
Price€59 / €249 excl. VAT/monthOn quote

Source: IgnitionRAG product facts published on ignitionrag.com, October 2026.

Document sources plug in through connectors: web scraping, SharePoint, Azure Tables, Azure Blob, S3 and MinIO, GitHub, Hugging Face.

The step everyone skips: who is responsible for what

The first topic in a white-label resale is contractual. Your client hands over documents. You have them processed by a platform. That platform relies on its own providers. Each link has a role under the GDPR.

In the most common setup, your client decides the purposes: it is the controller. Your agency or ESN processes on its behalf: it is a processor. The platform vendor, when it hosts the service, becomes a sub-processor. This qualification must be checked case by case with your DPO: it depends on who decides the purposes and the means.

The European Data Protection Board (EDPB) Guidelines 07/2020 set out two obligations that shape this setup:

  1. Prior written authorisation. The processor may not engage another processor without the controller's prior written authorisation, specific or general. With a general authorisation, it must actively inform the controller of any change and give it the opportunity to object (§128).
  2. Same obligations down the chain. The contract with the sub-processor must impose the same data protection obligations, including the possibility of audits (§129).

In practice: the platform's provider list must appear in your client contract, and you must be able to announce each change.

The trap to avoid: leaks from one client to another

An ESN serving ten clients on one platform creates a risk OWASP describes precisely. LLM08:2025 ("Vector and Embedding Weaknesses") lists cross-context information leaks. In a multi-tenant environment where several groups share the same vector database, embeddings from one group may be retrieved in response to another group's queries.

The OWASP "RAG Security Cheat Sheet" gives the countermeasures:

  • separate namespaces, collections or indices per tenant;
  • filtering enforced at query time, before similarity results are returned;
  • cross-tenant test queries, replayed regularly, that must return zero out-of-scope results.

Two architectures meet this need. One instance per client gives the most readable isolation, at the cost of multiplied operations. A multi-organisation platform pools operations, provided isolation is verified for every new client. IgnitionRAG's privacy policy describes multi-tenant data isolation per organisation. Still replay your own test queries between two organisations before onboarding a second client: OWASP recommends it, and it is your responsibility as a processor.

BYOK: who pays for the model, and on which account

BYOK means "bring your own keys": calls to the language model use API keys that belong to the agency or its client.

In IgnitionRAG, BYOK is available with a licence and does not exist on SaaS. The Solo and Team SaaS plans run on platform-managed AI with credits included: no key to provide. BYOK also exists at other vendors. The choice mostly comes down to three points:

QuestionWhy it matters
Which account are model calls billed to?Your margin and client re-billing depend on it
Does the model account belong to the ESN or the end client?It determines who sees the model provider's usage logs
Where does the chosen model run?Passages retrieved from documents are sent there with every request

With the end client's keys, the model invoice and contract stay with the client. With the ESN's keys, you carry that cost and re-bill it.

Where to deploy: at the client or with the vendor

The Agency/ESN licence deploys on the client's infrastructure (Azure, AWS or Google Cloud) or with IgnitionAI.

CriterionIn the client's cloudWith IgnitionAI
Cloud contracts and security auditThe client's, already in placeTo be added to the client's vendor file
Subprocessing chain to declareHosting covered by the client's contractsIgnitionAI is in principle added to the subprocessing chain
OperationsSplit between ESN, client and vendorInfrastructure run by the vendor
Typical caseLarge account with an audited cloudSME client without a cloud team

Our guide France-hosted RAG details the questions on applicable law and model location to ask for each option.

Decision grid

Your situationApproach to study
You want to evaluate the product firstBook a demo first (planned trial: 14 days, 100 credits, no payment card), then scope the licence before the first client
You serve several SMEs with no cloud teamAgency/ESN licence hosted by IgnitionAI, one organisation per client, isolation tests at every onboarding
Your client requires its Azure, AWS or GCP cloudLicence deployed in its cloud, with its own model keys
Your client handles highly sensitive documentsInstance on the client's infrastructure (Azure, AWS or GCP), DPO and CISO review before ingestion

Grid basis: EDPB Guidelines 07/2020 and the OWASP publications cited below. Mapping situations to approaches is our reading and is not legal advice.

What to prepare before the first client

  1. A processing agreement listing sub-processors and the notification procedure for changes.
  2. The deployment location, client by client.
  3. The account holding the model keys, and the re-billing rule.
  4. A set of test queries between two organisations, replayed for every new client.
  5. A withdrawal procedure: what happens to documents, index and traces when a client leaves?

Methodology and sources

Regulatory sources

Technical sources

IgnitionAI estimates

None. The Agency/ESN licence price is quoted on request.

Limits

This guide does not legally qualify your setup. Controller and processor roles are determined case by case, with your DPO.

Last source review

22 October 2026.

Get started

To scope an Agency/ESN licence (white label, BYOK, deployment location), contact us from the pricing page. The SaaS trial includes 14 days and 100 credits, without a payment card.

FAQ

Is white-labelling available on the Team plan?

No. White-labelling (colour, name and logo editable in the admin console) is sold with the Agency/ESN licence.

How much does the Agency/ESN licence cost?

The price is quoted on request.

Can we deploy at our client on Azure, AWS or Google Cloud?

Yes. The Agency/ESN licence deploys on the client's infrastructure (Azure, AWS or GCP) or with IgnitionAI, with BYOK available.

Is BYOK possible on SaaS?

No. The Solo and Team plans include platform-managed AI with credits included. BYOK is reserved for licences.

Which document sources can be connected?

Web scraping, SharePoint, Azure Tables, Azure Blob, S3 and MinIO, GitHub and Hugging Face.

Deploy AI agents that work for your company.

Connect authorized knowledge and tools, then put an agent to work where your team needs it.