GUI, API & Distribution
Unified entry points for users, builders and enterprise delivery.
Platform Move from a collection of capabilities
Move from a collection of capabilities
to a system that keeps working
Build, test, deploy, operate and govern around the same agent identities, assets and boundaries.
Platform capabilities vary by client version, runtime environment and connection configuration.
Cover the full digital-worker lifecycle
Each layer solves a long-term operating problem and provides a stable boundary for the layer above.
Governance Control Plane
Organization, permission, approval, audit, cost, recovery and lifecycle.
Sandbox & Connections
Connect models, files, browsers, desktops and business systems within scope.
AgentFS
Store identity, rules, memory, Skills, Workflows, receipts and versions.
Multi-Agent Runtime
Independent work, parallelism, delegation, handoff, review and recovery.
The difference is not agent count — it is whether durable work subjects exist
A persistent agent is not just a worker thread inside one call. It keeps an identity, assets and accountability boundary.
| Dimension | Temporary Worker | Persistent Agent |
|---|---|---|
| Identity | Created around the task | Stable role and durable responsibility |
| Assets | Returns output to the parent task | Own AgentFS and accumulating assets |
| Collaboration | Splits work and returns | Messaging, delegation, handoff, teams and review |
| Permission | Inherits current task scope | Own boundary capped by parent authority |
| Organization | Temporary hierarchy | Persistent teams, supervisors, members and workspaces |
Seven stages from intake to experience return
See how information enters the system, gets executed, governed and turned into reusable experience.
- 01
Intake & Perception
- Input
- Files, messages, events and business material
- Output
- Structured task input
- 02
Understand & Identify
- Input
- Context, identity and goals
- Output
- Constraints, gaps and acceptance criteria
- 03
Plan & Orchestrate
- Input
- Tasks and available members
- Output
- Steps, assignments and dependencies
- 04
Knowledge & Memory
- Input
- AgentFS, sources and experience
- Output
- Task-relevant context
- 05
Execute & Connect
- Input
- Tools, MCP, files, browser and desktop
- Output
- Intermediate artifacts and actions
- 06
Govern & Evaluate
- Input
- Permissions, policy, approvals and checkpoints
- Output
- Authorization and quality decisions
- 07
Deliver & Return
- Input
- Results, receipts and human feedback
- Output
- Deliverables and reusable experience
Turn expertise into manageable digital-worker assets
File-based does not mean dumping content into folders. It gives identity, knowledge, capabilities and governance explicit homes.
Identity & Rules
- Persona
- Principles
- Role boundaries
- Team rules
Knowledge & Experience
- Long-term memory
- Relationship memory
- Business sources
- Experience entries
Capabilities & Process
- Skill
- Tool
- MCP
- Workflow
Governance & Improvement
- Versions
- Evaluation
- Receipts
- Recovery records
InspectReviewMigrateVersionRecoverDeliver
DesireCore
Start with one agent, then build a team that keeps working
Experience the full product first, then evaluate platform and enterprise boundaries around your work.
