Skip to main content

Console V2 — Research Archive

:::warning Status: Research-only · Prototype archived Console V2 (internally codenamed Weave) was an exploration of a general-purpose, multi-tenant process-centric agentic platform for executing business workflows composed of versioned step graphs, modular skills, registered tools, and human governance.

Public commercial product development has been discontinued. This documentation is retained as a technical research archive describing the system's architecture, process domain, design decisions, prototype implementation, and empirical findings.

The documented architecture includes a combination of implemented components, validated subsystem designs, and research records. Console V2 is not offered as a production SaaS service. This archive does not imply ongoing maintenance, deployment support, or a commitment to future releases. :::

The research and architecture whitepaper is published by Pluggedspace:
📄 Whitepaper: Pluggedspace Console Enterprise Agentic OS Whitepaper

1. Project Overview: The Process-Centric Paradigm​

Console V2 investigated a fundamental architectural pivot in enterprise agentic computing: moving from conversational agent personas to versioned, process-centric execution graphs.

In early agent platforms (Console V1), individual "vertical agents" (such as a marketing agent, a business finder, or a finance auditor) were treated as distinct persona silos. This model proved brittle for enterprise operations, resulting in duplicated tool integrations, unpredictable autonomy, and governance bottlenecks.

In Console V2 (Weave), the primary unit of work became the Process—a tenant-scoped, versioned directed acyclic graph (DAG) of steps executed as a ProcessRun. Rather than relying on open-ended agents to decide their own operational trajectory, a process defines the business workflow deterministically, invoking agents, modular skills, and tools as capabilities at individual steps.

┌────────────────────────────────────────────────────────────────────────┐
│                   CONSOLE V2 (WEAVE) RUNTIME MESH                      │
├────────────────────────────────────────────────────────────────────────┤
│                           PROCESS DOMAIN                               │
│   Process (UUID) ──► ProcessVersion (Immutable) ──► ProcessStep (DAG)  │
│          ▲                        ▲                        ▲           │
│   Trigger (Webhook/Event)   ExecutionContext      Outcome / Terminal   │
├────────────────────────────────────────────────────────────────────────┤
│                         EXECUTION RUNTIME                              │
│   Planner / TaskPlan  │  ToolRunner  │  SafetyLayer  │  PolicyEngine   │
│   ModelRouter (LLM)   │  MemoryBus   │  SessionMgr   │  WorkspaceMgr   │
├───────────────────┬────────────────────┬───────────────────────────────┤
│  Tool Bus         │  Skill Registry    │  BrainBox Memory              │
│  29 auto-discovered│  ~95 domain skills │  4-layer substrate: events,   │
│  BaseTool schemas │  (finance, risk,   │  384-dim pgvector, KG, and    │
│  & file sandbox   │   growth, data)    │  inference (pluggable provider)│
├───────────────────┴────────────────────┴───────────────────────────────┤
│                         GOVERNANCE & CHANNELS                          │
│   ApprovalQueue (Manual / Hybrid / Autonomous) │ Monica Support Surface │
└────────────────────────────────────────────────────────────────────────┘

2. Key Architectural Pillars​

  1. The Process Domain: A formal state machine governing runs (draft → running → (waiting | paused) → completed / failed / cancelled) with 20 distinct step types (tools, skills, agent invocations, human approvals, conditions, branches, loops, and parallel forks).
  2. Decoupled Capability Substrates:
    • Tool Bus: 29 registered tools with explicit permission scopes, parameter validation, risk levels, and sandboxed output under MEDIA_ROOT/generated/.
    • Skill Registry: ~95 reusable domain skills. Capabilities originally siloed in separate vertical apps (finance, governance, maintenance, risk detection) were consolidated into first-class skills.
  3. Pluggable BrainBox Memory: A 4-layer memory substrate (event store, 384-dimensional pgvector semantic memory, knowledge graph, and inference layer) decoupled via a MemoryProvider so external memory failures never crash active process runs.
  4. Out-of-Band Governance: An ApprovalQueue supporting manual, hybrid, and autonomous operation, enforced by a deterministic PolicyRuleModel using glob-matched action rules and rate limits.
  5. Monica Omnichannel Surface: The only surviving customer-facing vertical—providing FAQ RAG, dispute handling, appointment scheduling, and a public web chat widget locked to verified origin allow-lists.

3. Scope & Research Limitations​

When reviewing this archive, note the operational context:

  • Implemented & Validated Subsystems: Process state machine, DAG execution planner, 29 registered tools, ~95 skill implementations, fail-closed multi-tenancy, BrainBox vector memory provider, and the Monica omnichannel support widget.
  • Retired & Decommissioned Surfaces: In V2.6, vertical agent personas (marketer, businessfinder, newsletter, finance, governance, maintenance) and the agent marketplace were retired. Tenant agent authoring was replaced by process composition (/api/processes/compose/).
  • Commercial Service Status: Public SaaS endpoints (agents.pluggedspace.org), multi-tenant subscriptions, and commercial support have been discontinued.

4. Archive Navigation​

Explore the technical records and design specifications preserved in this archive:

TopicDescription
System ArchitectureFull breakdown of the Process Domain, ExecutionRuntime, Tool Bus, Skills, BrainBox, and Governance.
Design DecisionsWhy vertical agent personas and the marketplace were retired in favor of process graphs and modular skills.
Prototype & EvaluationImplemented data models, test coverage, empirical benchmarks, and known limitations.
Research Paper & CitationCanonical whitepaper reference, publication abstract, version history, and BibTeX citation.
Prototype WalkthroughStep-by-step walkthrough of the Process Composer, step configurations, and operations notes.

5. Citation & Reference​

If you reference the Console V2 (Weave) architecture or empirical findings in academic research or technical publications, please cite:

@misc{pluggedspace2026consolev2,
  title={Pluggedspace Console: An Enterprise Agentic Operating System and Runtime Architecture},
  author={{Pluggedspace Research Team}},
  year={2026},
  howpublished={\url{https://pluggedspace.org/research/Pluggedspace-Console-enterprise-agentic-os-whitepaper}},
  note={Research Archive and Whitepaper}
}