Skip to main content
Djinny

The best AI models, without depending on them

Djinny brings Claude, GPT, Gemini, Mistral and open models into one tool, connected to your documents and managed by your IT team. You get the best each provider offers while keeping your freedom to choose.

Djinny answers using two internal documents, masks names in a request, then prepares a PDF brief.

  1. Documents
  2. Anonymisation
  3. PDF brief
What is our procedure for onboarding a new supplier?
ClaudeAuto
Four steps1: the supplier form, a quality assessment scoring at least 12/202, checks of the company registration and URSSAF certificates, then approval by the Purchasing manager1.
Sources 21Supplier onboarding procedure.pdfSharePoint2Responsible purchasing charter.pdfSharePoint
Example conversation
Your choice of modelsClaude, GPT, Gemini, Mistral and open models
Hosted in Franceor installed on your infrastructure
Switch modelswithin the same tool, keeping your assistants and their instructions

Why Djinny

Your AI provider has become an architecture decision

A company chooses a model because it is the best for its use cases. It builds its assistants, automations and processes around it. A few months later, prices have changed, relative performance has too, and switching models means reworking much of what was built.

Djinny keeps your work independent of the models. Your assistants, prompts, documents and access rights are managed in Djinny. We use technology from Anthropic, OpenAI, Google and Mistral. We keep your freedom of choice in your hands.

Sovereignty does not mean doing everything yourself. It means being able to change your mind.

What your teams see

Watch the video

Screenshots from the application, using demonstration data.

The agent searches your documents, runs calculations in an isolated sandbox and produces Word, Excel, PowerPoint or PDF files to download. Every step stays visible.

  • Search your knowledge bases
  • Calculations and charts in Python, with visible steps
  • Notes, workbooks and presentations to download
The agent compares three offers and writes the Word briefing for management.

Djinny : Agent mode

Djinny in Agent mode compares three supplier offers in a table and provides a Word briefing for management.

The agent compares three offers and writes the Word briefing for management.

And what you need to use it across the company

Expert modes

A business assistant with its own model, instructions and sources. Change the model in the settings and keep the assistant.

Company accounts

Sign in with Active Directory, LDAP or ADFS. Teams use their company account, with no API keys to distribute.

Groups and permissions

Models, services and libraries made available by team. Each person inherits their group’s permissions.

MCP connectors

Your business tools connected to agents within the limits set by IT. Every call is logged.

OpenAI-compatible API

Your own applications go through Djinny and use the same models under the same rules.

Getting started

A built-in tutorial guides each new user through their first tasks.

Help your teams get started

Deployment

With us in France. Or on your infrastructure

Hosted in France

Operated by Magic LEMP, with nothing to install on your infrastructure. Your IT team chooses the available models and the destinations allowed for requests.

On your infrastructure

Djinny and open models in your data centre or cloud. Processing stays on your infrastructure when using local models. Calls to external models follow the destinations authorised by your IT team.

Disconnected

An installation with no internet access for the most sensitive environments, using local models and suitable computing resources.

Hosting and security

Djinny in one minute

Download the video

Application demonstration in French, 1 min 12. Fictional data.

How do teams use Djinny?

Employees have an interface to ask questions and work with their documents. Djinny also acts as a gateway to multiple AI models for your applications.

What does request anonymisation do?

It handles identifying details in requests before they are sent to models, according to the selected rules. Together we define which data is handled, who can access it and where it is sent.

How can we monitor and optimise costs?

Djinny shows usage and costs to help choose models suited to each task. Savings depend on your volumes, models and selected rules.

Who decides where requests are sent?

Your organisation chooses the available models and permitted destinations. We define this scope with your IT department before deployment.

Does your AI use depend on a single provider? Let’s talk

Thirty minutes to see Djinny on your use cases and understand what it would take to switch models tomorrow.

A demonstrationusing your use cases and security requirements
A pilota limited scope and agreed indicators, in under six weeks
A deploymenthosted in France or on your infrastructure, at your own pace
Djinny: an AI gateway for your teams | Magic LEMP