aram Email us

Private AI for every employee, controlled by IT.

One AI assistant for your whole company, on your infrastructure. Your data never trains a model.

  • Model routing
  • Budgets by role
  • PII redaction
  • Access by role
  • Audit log
  • MCP gateway
Email us Estimate your costs
Your infrastructure, or colocated infrastructure
People
EmployeesAgentsYour apps
Workspace
ChatAssistantsAgent builder
Control plane
Sign-in and rolesBudgetsModel routingPII redactionAudit logMCP gateway
Your data
SharePointDriveSlackJira
Models on your infrastructure
QwenLlamaMistralYour own
Only masked requests leave
Frontier APIs
ClaudeGPTGemini

No training, no retention

1Priya in Finance asks about vendor contracts.
chat.yourcompany.com
  • AutoPicks the best-value model
  • Claude SonnetAnthropic
  • GPTOpenAI
  • GeminiGoogle
  • Mistral, on your serversOpen model
  • Qwen, on your serversOpen model
  • Llama, on your serversOpen model
  • Your own modelSelf-hosted
Not used for training
ToolsCompany filesWeb searchCode interpreterJiraSalesforce

Models

  • Claude
  • GPT
  • Gemini
  • Mistral
  • Qwen
  • Llama
  • DeepSeek
  • Your own models

Data and tools

  • Google Drive
  • SharePoint
  • Confluence
  • Slack
  • Jira
  • Salesforce
  • Web search
  • Code interpreter
  • Any MCP server

Your employees already use AI. IT can't see it.

  • People paste contracts, customer records and code into personal AI accounts.
  • Seat licences cost the same whether someone uses AI every day or never.
  • Each vendor ties you to its own models, its own regions and its own terms.

aram replaces that with one approved assistant for everyone. You see who uses it, what it costs and what data it touches.

What happens when someone asks a question

  1. They ask in aramSigned in with your company login, Microsoft Entra ID, Google Workspace or Okta.
  2. aram checksTheir role, their team's budget, which files they may see, and removes personal data if you turn that on.
  3. aram picks the modelThe cheapest model that handles this kind of question well: an open model for a summary, a frontier model for hard analysis. People can still choose one themselves.
  4. The model answersClaude, GPT, Gemini or Mistral under no-training terms, an open model such as Qwen or Llama running on your infrastructure, or your own.

Every request is logged: who asked, which model answered, what it cost.

Estimate your model costs

Pick what your company uses AI for and roughly how much. You'll see what it costs if everything runs on one frontier model, and what it costs when aram's automatic routing sends work that open models handle well to them.

Prices and assumptions used

    Runs where your data is allowed to be

    On your infrastructure

    We deploy aram in your AWS, Azure, GCP or on-premise environment and run it for you. Prompts and documents never leave it.

    On colocated infrastructure

    Dedicated infrastructure in a colocation data centre, run for you by us. Nothing to install, and updates arrive without work on your side.

    With the models you choose

    Claude, GPT, Gemini and Mistral under no-training terms. Open models such as Qwen, Llama and DeepSeek running on your infrastructure. Or your own models and provider accounts.

    aram is built on open-source software, so the setup we run for you is never locked to us.

    You pay for use, not seats

    We share prices on the first call. They depend on where aram runs and how many people use it.

    Active users
    A monthly fee for each person who used aram that month. Everyone else costs nothing, so you can give it to the whole company.
    Active agents
    A monthly fee for each agent that ran that month.
    Model usage
    Billed at what the model provider charges. No markup.
    Data protection checks
    Optional. Personal-data masking and policy checks, billed per checked request.
    Onboarding
    A one-time fee for setup, company login, roles, data connections and training your teams.

    Who aram is for

    A good fit if

    • You don't want your company's data exposed to AI model providers.
    • Not all your employees need the latest, most expensive models.
    • You have an IT team and hundreds or thousands of employees.
    • People already use AI at work, on their own accounts.
    • You use Google Workspace, or a mix of Microsoft and other tools.
    • You want more than one AI vendor, and the freedom to switch.

    Not for you if

    • You're an individual or freelancer. aram has no personal plans.
    • Your company runs entirely on Microsoft 365 and is happy with Copilot. It's built into the tools you already use, so stay with it.

    How a pilot works

    1. A 30-minute callWe learn how your company works and tell you plainly whether aram fits.
    2. SetupCompany login, roles, budgets, the models you choose, and your first data sources.
    3. 30 to 60 days, whole companyNot a small trial group, because adoption is what you're testing.
    4. The reportHow many people used it and what it cost compared with seats. Then you decide.

    Tell us about your company

    Email us your company size, your country and the AI tools people use today. We'll reply and tell you honestly whether aram is a fit.

    Email us

    or write to