Plane map¶
Target
This page describes intended behaviour or structure. No implementation is claimed unless a section is explicitly marked Current implementation or Evidence.
Restructured (2026-10)
The site was reorganized from twelve platforms to the planes below (user decision D2, 2026-10-08; executed 2026-10-09). Every former platforms/ address redirects to its plane or solution-pack page; the page-by-page mapping is in the review folder's ia-migration-map.md. The earlier five-plane grouping (Experience, Control, Work, Delivery & Runtime, Cross-cutting) is superseded by this page.
The Factory is an organizational AI operating and control platform. Its responsibilities are grouped into planes. A plane is a logical boundary: it names what must be owned, decided and observed. It does not prescribe a deployable service; the initial physical shape can be a modular control-plane API and a runtime worker over existing persistence, messaging and identity infrastructure.
flowchart TB
EXP["Experience plane<br/>web console · IDE · CLI · SDK · API · chat channels"]
CTL["Control plane<br/>tenants · projects · published assets · model profiles · policies · budgets · run authority"]
AGT["Agent & Orchestration runtime<br/>planner · executor · verifier · supervisor · HITL · checkpoints · recovery"]
MDL["Model & Compute<br/>model gateway · capability router · providers · compute pools"]
KNW["Knowledge & Memory<br/>documents · code · vector · graph · provenance · memory · context engineering"]
CAP["Capability & Tools<br/>skills · MCP · tools · connectors · tool registry · tool policies"]
EXE["Execution<br/>sandboxes · workspaces · shell/files/git/browser · pipelines · environments"]
DAT["Data & Event<br/>relational · cache · vector · graph · blob · event bus · audit store"]
TRS["Trust & Operations<br/>security · policy · evals · audit · observability · FinOps · reliability"]
EXP --> CTL --> AGT
AGT --> MDL
AGT --> KNW
AGT --> CAP
CAP --> EXE
MDL --> DAT
KNW --> DAT
EXE --> DAT
TRS -.-> EXP
TRS -.-> CTL
TRS -.-> AGT
TRS -.-> EXE
Planes and their current documentation¶
| Plane | Owns | Existing pages that migrate here | Status today |
|---|---|---|---|
| Experience | Human surfaces: console, review queues, run inspection, admin; IDE, CLI, SDK, API and chat channel entry points | Factory Studio and the ui.md page of every platform; Studio modules |
Prototypes and design pages; no console in code |
| Control | Tenants, projects, environments; publication of immutable agent/skill/tool/workflow releases; model profiles; policies and budgets; authoritative run, step, attempt, lease and approval state | Control Plane; registries from Agent Mesh; Multi-tenancy; project lifecycle pages | Design only; see ADR-0015 |
| Agent & Orchestration runtime | Executing a run against a pinned release: planning, steps, attempts, verification, supervision, human-in-the-loop, checkpointing and recovery | Agent Mesh runtime pages; Agent execution flow; Orchestration layer; Runtime failure and recovery | Lifecycle library and scaffold host exist; nothing durable |
| Model & Compute | Logical model profiles, capability and privacy routing, provider adapters, usage receipts, compute pools for model serving | Model access sections of Agent Mesh; new pages to be written | Four provider adapters exist; no profiles, routing or receipts |
| Knowledge & Memory | Approved sources, ingestion, vector and full-text retrieval, knowledge graph, provenance, memory lifecycle, context engineering | Knowledge Platform; Knowledge and memory system | Chunking, embedding and search samples; no scoped retrieval; see ADR-0013 |
| Capability & Tools | Skill definitions, MCP servers, executable tools, connectors, tool registry, tool policies, capability distribution | Integration Platform; Marketplace; MCP governance | MCP hosting and governance library exist; AgentKit holds skills |
| Execution | Per-run sandboxes and workspaces, shell/file/Git/browser execution, build and delivery pipelines, environment provisioning | DevOps & GitOps; Runtime & Cloud | Templates and CI pipelines exist; no Factory sandbox |
| Data & Event | Store profiles (relational, cache, vector, graph, blob), event bus, envelopes, audit store, catalogs of events, storage and aggregates | Event-Driven Architecture; Data Architecture; Event, Storage and Aggregate root catalogs | Messaging port and adapters exist; see ADR-0012, ADR-0014 |
| Trust & Operations | Security, policy engine, approvals policy, evaluation, audit, observability, FinOps, reliability, compliance | Governance, Security & Compliance; Observability & Feedback; Security, Observability and Resilience architecture pages | OpenTelemetry and Audit.NET integrations exist; no policy engine |
| Solution Packs | Software Factory: engineering agents, workflows, blueprints, coordinators, templates and libraries. Generated SaaS: the products produced and their own tenants, stores and operations | Agent system, Workflows, Template & Library Platform, Templates; Generated SaaS Product Platform | Richest existing content; templates real, generation pipeline not |
Mapping the twelve platforms¶
| 2026 platform | Plane(s) |
|---|---|
| Factory Studio | Experience |
| Control Plane | Control; its policy and audit rows move to Trust & Operations |
| Agent Mesh | Agent & Orchestration runtime; its registry pages move to Control; model access to Model & Compute |
| Knowledge Platform | Knowledge & Memory |
| Template & Library Platform | Solution Pack: Software Factory |
| Marketplace | Capability & Tools (its UI to Experience) |
| DevOps & GitOps | Execution (its workflows page to the Software Factory pack) |
| Runtime & Cloud | Execution |
| Observability & Feedback | Trust & Operations |
| Governance, Security & Compliance | Trust & Operations; approval instances to Control |
| Integration Platform | Capability & Tools |
| Generated SaaS Product Platform | Solution Pack: Generated SaaS |
Where a name appears in two platforms today (agent and skill registries, approval, policy and audit services, IaC provisioning), the ownership map assigns one authoritative owner: publication to Control, release loading to the Agent runtime; approval instances to Control, approval policy and audit to Trust & Operations; provisioning to Execution.
Cross-cutting rules that apply to every plane¶
- Trusted scope. Tenant and project identity come from the authenticated context under canonical ADR-0100/0101; a selection in the console grants nothing. See ADR-0015.
- One execution port. REST, MCP, queue, workflow and IDE entry points invoke the same application port; raw framework calls do not bypass lifecycle, policy or accounting.
- Immutable releases. A run pins the exact agent, skill, tool, workflow, policy and model-profile versions it started with.
- Explicit uncertainty. Cancellation, timeouts and lost replies leave durable, visible state; nothing is reported as done because a message was sent.
- Evidence before claims. A page may describe a target fully, but it may not claim implementation, deployment or guarantees without dated evidence.
Related¶
- AI Software Factory Overview — the twelve-platform generation of the target design
- Ecosystem Map
- Architecture Decision Records
- Capability dependencies