Datastreams
    How Datastreams works

    Tell us what should happen.
    We turn it into a running operation.

    Start with the information that arrives, the rules your organisation follows and the result you need. Datastreams turns that description into a controlled data process and keeps the underlying runtime working.

    Start with one recurring process. Your team does not need to assemble a data platform first.

    One understandable description

    Five questions describe the complete business process.

    Business owners can start with what comes in, what must happen and what result they need. The technical model sits underneath that conversation.

    01

    What comes in

    The order, payment, customer signal, document or other information that starts the process.

    02

    What it means

    The customer, product, permission and current business situation needed to understand it.

    03

    What must happen

    The checks, business rules, exceptions and decisions that always need to be applied.

    04

    Where the result goes

    The system, person or approved AI that receives the answer or may take the next action.

    05

    What stays visible

    The active version, applied rule, result and information needed to explain what happened.

    The technology underneath

    DimML keeps the agreed operation readable and versioned.

    This simplified example shows the information that belongs together. It is not published as formal DimML or CLI syntax until the current language documentation has been verified.

    stream:
      source: customer_activity
      contract: activity.v1
      context:
        - customer_permissions
        - current_session
      rules:
        - allow_permitted_purpose
        - reject_invalid_identity
      destinations:
        - operations_store
        - approved_agent
      evidence:
        retain: decision_trace

    AI can help prepare

    Use AI for the conversation, not as an unchecked production authority.

    AI can help turn examples and business rules into a proposal that people can review. Your organisation still decides what is correct and what may go live.

    Explain the need

    Describe the process in normal language and provide the examples and exceptions that matter.

    Review the proposal

    Make the incoming information, rules, destinations and intended changes visible before publication.

    Validate the definition

    Use the DimML toolchain to check structure and readability before remote changes are made.

    Approve and publish

    A responsible person approves the proposal before it is published to the selected runtime.

    One model, more responsibility

    A personal-data stream is not a different platform.

    It uses the same source-to-outcome definition. The regulated operation simply makes more context, policy and evidence explicit.

    Simple data delivery

    source: orders
    contract: order.v1
    destination: operations

    Personal and regulated data

    source: customer_activity
    contract: activity.v1
    context: [identity, consent, purpose]
    processing_basis: approved_legal_basis
    rules: [permitted_use, retention]
    destination: permitted_operations
    evidence: decision_trace

    Conceptual configuration model. Exact fields and CLI commands are published only after validation against the current runtime implementation.

    Technical background: DimML

    Runtime insight per recipe

    See what is running, why it may run and what it produced.

    Each active configuration keeps operational and governance insight at recipe level, even when hundreds of different data operations share the runtime.

    01

    Active version, publisher and approval evidence

    02

    Declared purpose and approved legal basis

    03

    Accepted, rejected, retried and recovered events

    04

    Rules, policy outcomes and permitted actions

    05

    Processing time, active state and delivery behaviour

    06

    Destinations, provider adapters and resulting evidence

    The accountable organisation determines and approves the processing purpose and legal basis. Datastreams makes the declared basis and resulting runtime decisions inspectable; it does not provide an automated legal conclusion.

    A visible route to production

    Prepare, check and publish without an invisible shortcut.

    The installed DimML CLI already supports local readiness checks, linting, validation, dry-run planning, publication, verification and runtime inspection. Exact examples will follow with the formal documentation.

    1. 01

      Prepare

      Turn the agreed business process into a readable, versioned proposal.

    2. 02

      Validate

      Check the structure and readability locally before anything is published.

    3. 03

      Preview

      Inspect the files and planned changes without writing them to the selected runtime target.

    4. 04

      Approve

      Let the responsible person decide whether the reviewed proposal may go live.

    5. 05

      Publish

      Send the approved configuration to the selected runtime environment.

    6. 06

      Verify

      Confirm that the published files are visible and inspect runtime status and logs.

    Designed for people and agents

    An agent can use the same checked route as a human operator.

    The DimML CLI is designed for both people and agents. It provides a direct operating route without making the AI model the owner of production decisions.

    Read and inspect

    List installed files and inspect runtime visibility, status and logs.

    Prepare and validate

    Create a versioned proposal, lint it, validate it and preview planned file changes.

    Publish after approval

    Keep proposal generation separate from the credentials and decision required to publish.

    Where to begin

    Start with one recurring process that matters every day.

    These are examples of business processes that can be described first. Verified DimML examples will be added when the formal documentation is available.

    Customer activity

    Validate interaction events, attach consent and current context, then route analytics and permitted actions.

    Transactions and risk

    Evaluate a payment or order while it moves and intervene before it becomes a delayed warehouse record.

    Operational exceptions

    Detect incomplete, late or inconsistent events and route them to a system, person or approved agent.

    Documents and messages

    Extract, validate and enrich incoming information before it enters operational systems.

    Devices and telemetry

    Apply contracts, state and operating rules to high-volume machine events in real time.

    Multi-destination delivery

    Deliver one governed event to operational, analytical, archival and agent consumers without rebuilding the stream.

    Provider replacement

    Validate and approve a compatible source, identity, trust, payment or destination adapter while preserving the stream contract and audit history.

    Turn one recurring business process into a controlled data operation.

    Bring the information that starts it, the rules that matter and the result you need. We will determine what can already be verified and what still needs product documentation.