Identity & Role
Stable identity, responsibility and boundaries
Enterprise Bring every agent into your
Bring every agent into your
permission, knowledge and accountability system
Manage people assignments, multi-level memory, tools, data, approvals, operations and audit across the full digital-worker lifecycle.
Enterprise console and centralized policy capabilities depend on the current release, deployment model and solution assessment.
The managed subject is the agent — not a single run
An enterprise agent is a durable digital worker. Runs, Checkpoints and receipts are its work records.
Enterprise Agent AgentFS
Organization & Owner
Department, team, accountable owner and service scope
Assets & Version
AgentFS, capability baseline and release state
Memory & Knowledge
Multiple scopes, depths and promotion boundaries
Tools & Data
Capabilities, directories, connectors, models and budgets
State & Work Records
Active work, exceptions, receipts, evaluation and audit
See effective boundaries from the organization down to one agent
Enterprise Agent
Published Financial Reporting Agent
Identity, role, owner, service scope, version and state
Scopes, access policy, promotion and provenance
Tools, Skills, MCP, directories and connectors
Templates, Human Gates, approvers and exceptions
Effective policy
Policy v12 Calculated from enterprise ceilings, department policy, user relationship and task context.
Answer precisely: who can ask which agent to do what
Enterprise ceilings flow downward; explicit denial wins; delegated authority is the intersection of caller, target agent and current task.
Who
User, group, department, role or agent
On what
Agent, memory, tool, data, model or work record
Can do what
Discover, use, read, write, delegate, approve or export
Under which conditions
Environment, origin, risk, time, budget and expiry
With what result
Allow, deny, require approval or return masked content
Grant a specific subject specific actions on specific memory levels of a specific agent
Policy combines ownership scope, content depth and action — never a single “memory access” toggle.
Enterprise Memory
Policies, terminology, knowledge standards and cross-department rules
Department / Team Memory
SOPs, project knowledge and shared experience
Agent Core Memory
General expertise and durable experience of the agent
User–Agent Private Memory
Personal corrections, working relationship and private experience
User Profile & Preferences
Cross-agent identity, preferences and privacy rules
Task Working Memory
Current material and unconfirmed intermediate conclusions
Content depth L0 L1 L2 Allowed actions
Summary
Know the topic and minimal conclusion
Key Points
Read important rules, facts and decisions
Full Detail
Access sources, examples and sensitive detail
DiscoverReadInject ContextAddCorrectPropose PromotionApprove PromotionRevokeExport
Example: Finance user × Financial Reporting Agent
Corporate finance policyRead L0/L1; L2 requires approval; export denied
Finance team memoryRead L0/L1/L2; may propose additions, cannot publish directly
Agent core memoryInject L0/L1; agent reads L2 only under policy
Personal private memoryUser can read, write, correct and delete; admins see metadata by default
Current task materialUsable in the task; delete, archive or propose promotion at completion
Memory policy must be enforced consistently across retrieval, context assembly, direct file access, search, export, cloning and automated memory work — not merely hidden in the admin UI.
Manage enterprise agents like durable job capabilities
- 01
Draft
Edit identity, rules, memory and capabilities
- 02
Test
Use only test users and sandbox data
- 03
Pending Approval
Freeze version for business and security review
- 04
Published
Serve authorized users in production
- 05
Suspended
Stop accepting work and handle active tasks
- 06
Retired
Disable use and retain records by policy
Keep low-risk work flowing and make high-risk responsibility explicit
Visible
Inspect agent state, active work, cost, exceptions and outcomes.
Controlled
Configure Human Gates, approval chains and temporary grants by risk.
Recoverable
Use retries, Checkpoints, pause and human takeover to handle failure.
Accountable
Record agent version, policy, memory access, approvals and deliverables.
A work receipt should answer
01Who initiated02Which enterprise agent03Which capability version04Which policies applied05Which memory levels were accessed06Which tools and data were used07Who confirmed critical actions08Deliverables and recovery records
Every deployment model must make trust boundaries explicit
Identity and organization source
AgentFS and memory location
Memory retrieval and context assembly location
Model request path and content boundary
Agent and background task runtime
Connector credential ownership
Policy, audit and receipt retention
Update, backup and incident responsibility
DesireCore Enterprise
Start with one real role and one controlled workflow
Define responsibility, memory boundaries, tools, data, approval points and acceptance evidence before expanding deployment.
This page presents the enterprise solution direction. Availability is confirmed against the current release and formal solution scope.