Business and operations
Ask what is happening now and prepare the next permitted action without waiting for a new dashboard, export or manual handover.
Faster decisions and less coordination work
Coding agents for governed data operations
Coding agents can work with your live business.
Only your rules can change it.
Connect Codex, Claude Code, Gemini CLI, GitHub Copilot CLI, Mistral Vibe or your own approved coding agent to current business context. The agent can use scoped MCP tools or invoke the Datastreams CLI; the runtime decides what may be read, changed or sent.
The agent can write and run code. The runtime still controls the data operation.
Approved coding clients
Client independentOpenAI Codex
Terminal coding agent
Claude Code
CLI and MCP client
Gemini CLI
Enterprise and API CLI
GitHub Copilot CLI
Terminal-native coding
Mistral Vibe
CLI coding agent
Your own agent
MCP, API or CLI
Datastreams runtime
One DimML-defined capability boundary
Current contextPermitted actionsExecution evidence
Why it matters and who it serves
Give every team the speed of AI without giving up operational control.
AI becomes useful when it can work with the business state that applies now and help move a process forward. Datastreams makes that possible through one reusable contract for context, permission, action and evidence.
Data, IT and product teams
Serve Codex, Claude Code, Gemini CLI, GitHub Copilot CLI, Mistral Vibe, applications and your own agents from one governed capability instead of maintaining a separate integration for every interface.
One contract and fewer point integrations
Risk, privacy and compliance
Keep identity, purpose, personal context, approvals and evidence in the runtime, where they can be enforced independently of a model or prompt.
Control and audit evidence by design
From conversation to controlled operation
Ask for the outcome. Keep the authority in your business.
Natural language makes the operation accessible. It does not replace the exact contract underneath. Every client reaches the same governed source, context, rules, actions and evidence.
- 01
Ask in business language
A person or agent asks for an answer or outcome without needing to know the underlying source, schema or endpoint.
- 02
Resolve current context
The runtime selects the permitted live state, source freshness, identity, purpose and business situation needed for that request.
- 03
Invoke a named operation
The AI calls a typed query or command exposed from the approved service contract. It does not receive unrestricted database access.
- 04
Validate, approve and evidence
DimML rules validate the request, enforce authority and approvals, execute the permitted action and retain its confirmed result.
One request, one traceable result
The model proposes meaning. The runtime executes the agreement.
A query returns only the permitted current view. A change must use a named, typed command. The model can vary while the operational checks remain explicit, versioned and inspectable.
Natural-language request
“Show the orders whose current delivery context needs service attention, then prepare the response allowed for each customer.”
DimML-defined operation
Resolve identity and role
Read current order context
Apply service eligibility rules
Require approval where configured
Send the permitted command
Retain acknowledgement and outcome
Changed data, context or rule versions may produce a different next result. Deterministic execution means the same accepted input, context snapshot and approved rule version follow the same rule-based path.
One interface, many business contexts
Move from asking about the business to operating it responsibly.
The same pattern applies wherever a current answer and a controlled next action belong together. Each implementation defines its own sources, roles, freshness, actions and approval thresholds.
Commerce
Ask which orders need attention, explain the current delivery context and invoke the permitted service response.
Finance
Explain a ledger exception, prepare a correction and deliver it to Exact, AFAS or another authorised endpoint after the required checks.
News and media
Query current audience context and publish an allowed segment or propensity signal to the next business service.
Connected services
Inspect a case, handover or operational state and perform only the action allowed for that actor and moment.
Use the interface that fits
MCP, CLI and APIs operate the same controlled service.
There is no separate AI copy of the business rules. People, scripts, applications and agents use the same definitions, permissions, validations and runtime evidence.
MCP for AI clients
Expose scoped tools to an MCP-capable client. Read-only and mutating operations remain distinguishable, and each call reaches the runtime under an authenticated authority boundary.
Datastreams CLI
Approved coding agents and technical teams can query current records, send governed business messages, validate definitions and inspect the same service lifecycle from a terminal. The runtime still validates every data operation against your rules.
APIs for applications
Portals, apps, services and agent workflows use the same permitted queries and actions without rebuilding the processing agreement in every interface.
Bring one AI question and the business action that should follow.
Start with one live question and one governed action. We map the context, command, deterministic checks, approval and result into one controlled service.