Understand your work OS

AI capabilities: context, preparation and execution

Understand what the OS can read, what it can prepare and what actually changes the work.

Read, prepare or execute: which result do you need?

The initial customer journey allows authorised offers to be browsed and an enquiry to be submitted. Other reads and actions described here depend on deployed tools, modules and permissions. An assistant reply does not by itself prove that an action was executed.

Workflow example

Read, prepare, execute: three different effects

  1. Read the customer’s available context. No message is sent.

  2. Prepare a reply with a clear recipient and purpose.

  3. Follow the confirmation policy of the action.

  4. Check the execution result before saying the work is done.

Fictional illustration. Available capabilities, sources and access depend on your environment; conditions are explained below.

On this page
  1. Analysis begins within an authorized scope
  2. An answer is not an action
  3. Execution flows are explicit
  4. Capacity and activation follow deployment
  5. Assess a useful answer
  6. Tools and prerequisites

Analysis begins within an authorized scope

A contact, call, message, appointment, report or mission supplies context only if the current identity can access it. The assistant works with the sources and modules available in the environment.

A missing connection, insufficient role or absent content must not be hidden behind a plausible answer.

An answer is not an action

Reading, organizing or explaining context does not mean a message was sent, an appointment created or a record modified.

Briefings and daily summaries read visible work. A preparation can suggest a next step without having executed it.

Execution flows are explicit

Action drafts can provide a preview and an action_id for confirmation. Missions have their own states and confirmation, rejection or archiving operations.

Some internal changes are directly possible after an explicit, authorized request. Behaviour depends on the tool and its policy; it cannot be reduced to a promise of systematic confirmation for every operation.

Capacity and activation follow deployment

Enabled modules, connections, volumes and operational needs determine capacity scope. Limits actually applied to an account or workspace remain applicable.

Capacity, volumes and applicable limits are defined according to enabled modules, usage and the agreed project scope.

Assess a useful answer

A useful answer identifies the context used, distinguishes facts from assumptions and indicates missing data. A useful action keeps its target, content, authorization and result.

Modules can evolve with the business. The possibility of a custom adaptation must not be presented as an already delivered function.

An answer, a proposal or a completed action?

StageExampleEffect
Read and analyseFind a customer’s unresolved objection.Returns authorized context. Does not send a message.
PrepareDraft the response and identify its recipient.Produces a proposal to review; it is not evidence of delivery.
ExecuteInvoke an authorized sending operation.Applies the specific tool’s access and confirmation contract; check the returned result and delivery status.

Operational reference

Modules, tools and access conditions

This reference describes MCP operations and their effects. The list exposed to your assistant depends on the deployed server, enabled modules and permissions. Reading, preparing and executing are distinct operations.

Signals and controlled actions

Adding a signal can enrich memory and prepare pending work without executing an outbound action. An action_id identifies prepared work; its execution and result must remain distinguishable from a read-only answer.

ToolFunction and effect
signals.searchSearch extracted signal titles and content by keyword.
signals.getRead an accessible signal, its participants and linked missions.
signals.addEnrich memory from a structured note or source; missing details can require clarification. No outbound action is executed.
signals.updateCorrect or enrich an existing signal without creating another one.
alefos.briefingRead a compact briefing across visible actionable work and recent context.
alefos.daily_summaryRead a summary for one local calendar day within the current scope.
actions.confirmExecute previously prepared work identified by action_id after confirmation.

The scenarios on this page illustrate workflows, not real customer data or guaranteed outcomes. Connections and capabilities must be enabled and verified for your environment. Official business systems retain their responsibilities.

Continue exploring

A capability of your company’s environment. Your team can use this capability from a compatible assistant through its configured MCP access. The deployment defines the enabled tools and permissions. Understand your AI connection ↗

Choose your next step.