Start a project →
Start a project →

AI Systems / xagents

Delegate the work.Keep the control.Delegate the work. Keep the control.

Xagents creates agents with bounded responsibilities that can analyse, decide and execute work using your business tools.

EVERY AGENT KNOWS WHAT IT IS RESPONSIBLE FOR, WHAT IT IS ALLOWED TO DO AND WHEN IT MUST STOP.
Start a project →
Operating boundary — the mandate
MandateReasonAct
VerifyCompleted
EscalateStop → human. Conditional, not a stage.

Inside the boundary: authorized context, tools and rules.
Outside: restricted actions and the human reviewer.

  1. Mandate
  2. Reason
  3. Act
  4. Verify → Completed
  5. Escalate → Person
  1. Mandate
  2. Reason
  3. Act
  4. Verify
  5. Escalate

02What an agent is

An agent is not a chatbot with access to tools.

An assistant can answer, summarize or draft. An operating agent receives an objective, uses context and tools, makes decisions inside clear boundaries and knows when the work is complete or must be handed to a person.

Assistant
  1. Question
  2. Response
Agent
  1. Objective
  2. Context
  3. Decision
  4. Tool
  5. Action
  6. Verification
  7. Complete
Human

Complete / Human — alternative outcomes after verification, not sequential stages

03When Xagents is useful

When the work requires judgment as well as sequence.

Fit conditionsSix situations
  1. 01
    Language

    Interpret documents, emails or conversations.

  2. 02
    Judgment

    The next action depends on context.

  3. 03
    Tools

    Query or update business systems.

  4. 04
    Multi-step work

    Several actions to reach the objective.

  5. 05
    Exceptions

    Not every case follows the same path.

  6. 06
    Ownership

    The system knows when the task is actually complete.

If deterministic logic solves the work better, we do not add unnecessary autonomy.

04Capability

From a bounded task to a coordinated crew.

  1. 01
    One mandate · one tool · review

    Task

    One concrete objective, few tools, explicit rules and frequent review.

  2. 02
    Context / memoryOne mandate · several tools · controlled actions

    Operator

    Several tools, context or memory and controlled actions.

  3. 03
    Separate mandates · shared state

    Crew

    Several specialized agents coordinated around shared state or context.

Task → Operator → Crew

Coordination grows. Permissions do not follow automatically.

05Boundaries

Autonomy starts by defining boundaries.

Agent mandate plate — field labels only

  1. Mandate
  2. Context
  3. Tools
  4. Rules
  5. Memory
CompletionEscalation → human
  1. 01
    Mandate

    What work the agent performs.

  2. 02
    Context

    What information it may use.

  3. 03
    Tools

    Which systems and actions it can access.

  4. 04
    Rules

    Which limits and policies it must respect.

  5. 05
    Memory

    Which state it needs to retain.

  6. 06
    Completion

    How it knows the work is finished.

  7. 07
    Escalation

    Which situations must be handed to a person.

06Control

More autonomy does not mean less control.

Three control paths — alternatives, not levels · Authority is fixed by the mandate and technical controls
Decision within mandate
Autocomplete
Approval
  1. prepare
  2. approve
  3. execute
Escalatestop → human

Approval is a deliberate control, not a failure. Coral marks intervention only.

The Xagents level describes the system's complexity. It does not grant permission to act. Actual authorization depends on the mandate and technical controls.

07Verification

We measure completed work, not agent activity.

Evaluation framework — no figures

  1. Work delegated
  2. Verified

Lime = the task actually completed, not activity produced.

  1. 01
    Verified completion.
  2. 02
    Correction rate.
  3. 03
    Escalation quality.
  4. 04
    Unauthorized actions.
  5. 05
    Review time.
  6. 06
    Cost per completed task.

08Composition

The agent decides. Other systems enable it to work.

One possible configuration — illustrativeXagents does not depend on the other families
  1. Xmindsupplies context
  2. Mandate
    Xagentsdecides within its boundary
  3. Xflowmay coordinate authorized actions

Business systems

09Observability

An agent in production needs observability.

We need to know what it decided, which tool it used, what result it obtained and when intervention was needed.

Audit schema — field names only

Conceptual. No run data.

  1. DecisionWhat it decided
  2. ToolWhich tool it used
  3. ResultWhat result it obtained
  4. Human interventionWhen intervention was needed

10 / Start a project

Every agent knows what it is responsible for, what it is allowed to do and when it must stop.

Start a project →
Bounded. Verified.

02 / What an agent is

An agent is not a chatbot with access to tools.

An assistant can answer, summarize or draft. An operating agent receives an objective, uses context and tools, makes decisions inside clear boundaries and knows when the work is complete or must be handed to a person.

Assistant

Question → Response.

Agent

Objective → Context → Decision → Tool → Action → Verification → Complete

/ Human.

03 / When Xagents is useful

When the work requires judgment as well as sequence.

  1. 01
    Language

    Interpret documents, emails or conversations.

  2. 02
    Judgment

    The next action depends on context.

  3. 03
    Tools

    Query or update business systems.

  4. 04
    Multi-step work

    Several actions to reach the objective.

  5. 05
    Exceptions

    Not every case follows the same path.

  6. 06
    Ownership

    The system knows when the task is actually complete.

    If deterministic logic solves the work better, we do not add unnecessary autonomy.

04 / Capability

From a bounded task to a coordinated crew.

  1. 01

    Task

    One concrete objective, few tools, explicit rules and frequent review.

  2. 02

    Operator

    Several tools, context or memory and controlled actions.

  3. 03

    Crew

    Several specialized agents coordinated around shared state or context.

Coordination grows. Permissions do not follow automatically.

05 / Boundaries

Autonomy starts by defining boundaries.

Agent mandate plate — field labels only

Mandate
What work the agent performs.
Context
What information it may use.
Tools
Which systems and actions it can access.
Rules
Which limits and policies it must respect.
Completion
When the work is verified as done.
Escalation
Stop → human, when out of mandate.

06 / Control

More autonomy does not mean less control.

Auto

Decision within mandate → complete.

Approval

Prepare → approve → execute.

Escalate

Stop → human.

The Xagents level describes the system's complexity. It does not grant permission to act. Actual authorization depends on the mandate and technical controls.

07 / Verification

We measure completed work, not agent activity.

Six dimensions
  1. 01
    Verified completion.
  2. 02
    Correction rate.
  3. 03
    Escalation quality.
  4. 04
    Unauthorized actions.
  5. 05
    Review time.
  6. 06
    Cost per completed task.

Lime = the task actually completed, not activity produced.

08 / Composition

The agent decides. Other systems enable it to work.

  1. Xmindsupplies context
  2. Xagentsdecides within its boundary
  3. Xflowmay coordinate authorized actions → Business systems

09 / Observability

An agent in production needs observability.

We need to know what it decided, which tool it used, what result it obtained and when intervention was needed.

  1. DecisionWhat it decided
  2. ToolWhich tool it used
  3. ResultWhat result it obtained
  4. Human interventionWhen intervention was needed

10 / Start a project

Every agent knows what it is responsible for, what it is allowed to do and when it must stop.

Start a project →