Skip to content

Current state

Evidence (dated)

This page records observations at a stated revision and date. Re-verify before relying on it; it is not a permanent contract.

This is the one page that says what is real. Everything else in the planes, packs and catalogs is the target architecture unless a section is marked Current implementation or Evidence. The facts below were observed in source on 2026-10-09 at the revisions listed; "exists" means the source was inspected at the stated scope, never that it is published, deployed, tested or verified unless that is stated. The site holds no running service: 1 page is labelled Current, 701 are Target, and every Target page will drift once implementation starts, which is why the rule at the end exists.

By repository and revision

Repository Revision What is there What it is not
ConnectSoft.Extensions.AI.Factory / .Abstractions / .Messaging / .Messaging.MassTransit 378ca89 / a85cc9b / 0d4767b / d4599ff Execution lifecycle (FactoryAgentBase: accept → prompt → run → validate → retry ≤ 3 → complete, escalate or fail), request/result/trace envelopes, a MassTransit publisher No run identity, tenant scope, persistence, checkpoint, cancellation stage or idempotency; pins drift between repositories
ConnectSoft.Extensions.AI, .AgentsFramework e02713e, 2d61ca4 IChatClient registrations for OpenAI, Azure OpenAI, Ollama, Azure AI Inference; embeddings; Semantic Kernel vector connectors (InMemory, Azure AI Search, PgVector, Qdrant, SQL Server); keyed agents, tools, RAG, file skills, DevUI No model profile, routing policy, usage receipt or Anthropic adapter; only InMemory is wired in the agent host
ConnectSoft.AI.SoftwareFactory.AgentTemplate (+ base-template) 9639e63 (bc6ac93) Agent host scaffold, 29 projects on .NET 10: sample agent, MCP tool wrapper, REST use case, Orleans grain, Hangfire jobs, MassTransit saga scaffold, retrieval sample, migrations, Audit.NET, OpenTelemetry Three entry paths, two bypass the lifecycle; retrieval not tenant-scoped; MCP unauthenticated by default
ConnectSoft.Extensions.ModelContextProtocol 8e7feef MCP hosting, optional authorization, Microsoft.AgentGovernance integration with policy paths, fail-closed and audit options Governance is opt-in
ConnectSoft.Extensions.MessagingModel / .Orleans / .MassTransit / .NServiceBus 29c8e77 / d623cfd / c811b07 / ac70b3a The messaging port (IEventBus, commands, events, handlers); Orleans outbox/inbox grains with NHibernate writer, SQL schema and 33 test files; MassTransit transports (options only); NServiceBus adapter README of the Orleans adapter is stale; profile evidence (contract tests per ADR-0012) not produced
ConnectSoft.Extensions.Saas.*, Identity and AuthorizationServer templates, ConnectSoft.Saas.MeteringTemplate 64c8082 and template repositories Tenant resolution per canonical ADR-0100, OpenIddict issuer per ADR-0101, transactional usage and quota metering Not bound into any Factory run, tool or retrieval scope
Template estate (BaseTemplate, Microservice, Worker, ApiGateway, Library, ApiLibrary, AISkillsLibrary, Platform, Blazor Shell, MFE and component templates) see the libraries and templates catalogs Real build tooling (Prepare-ExtendedTemplatePack, Invoke-TemplateCompose, Azure Artifacts pipelines); 43 templates, about 125 libraries inventoried (status shipping, stub, empty by the L1 rule) Generated-output validation and feed state not verified
ConnectSoft.Saas.TenantsTemplate / ProductsCatalog / Entitlements / Billing / Metering ed6fad0 / 89e5255 / ad6871f / 36c193f / a0561ef (2026-09-16) Descriptors, schemas and regenerate.ps1 per repository No script dispatches to a generator; the descriptor is an import format into blueprints since ADR-0025
DerechHaLev / Bereshit 2.0 (DerechHaLev.DocumentationSite 42af0c2) 42af0c2 Reference consumer of the template estate: gateway, authorization server, identity and six domain services on BaseTemplate; seven service projects reference the Orleans messaging adapter; approved design pack MSG1–MSG5 Its deployment and operation are not verified in this repository
ConnectSoft.AI.AgentKit (5238957), .Platform (bf69afb), .Agents (f600e3f), .OpenClaw (3ce3370), .Workforce (9d960d4) as listed Earlier, non-Factory experiments (D19): 100 agent definitions, 84 skills, 6 placeholder workflows, catalogs, IDE adapters, OpenClaw runtime configuration Reference input only; not Factory assets; three status vocabularies; 36 skills unreviewed
ConnectSoft.DesignPrototypes afc511b (2026-09-19) 41 prototype exports with fixture data and zero network calls; Factory prototype with 48 destinations (15 dedicated views); Workflow Studio prototype; design-system prompt and checklist Design evidence only; no Studio repository exists
This documentation site working tree, uncommitted since d8d43f40 The target architecture: 9 planes, 16 platforms, 42 bounded contexts, 191 service rows with typed ER and DDL, 100 agent specifications, 25 workflow definitions, 866 seeded skills, 25 ADRs, generated catalogs and traceability A design, not an inventory; see the rule below

The components the Factory is assembled from, each marked exists, extend or develop, are listed on the ecosystem composition page from model/components.yaml.

Capability matrix

Status values follow the original handoff §78: EXISTING, PARTIAL, MISSING, DUPLICATED, CONFLICTING, DEPRECATED, UNKNOWN.

Target capability Status Current evidence Gap to target
Agent and skill authoring with IDE adapters EXISTING (as an earlier experiment) AgentKit .ai sources, bindings, generated adapters, verifier script Review workflow and release gates are separate concerns; not Factory assets (D19)
Asset registry and immutable release resolution PARTIAL File catalogs in AgentKit and Platform; no release service No manifest, hash or closure publication; three status vocabularies
Desired-state control plane MISSING Design pages only No project, run or approval API exists
Run lifecycle and durable orchestration PARTIAL FactoryAgentBase lifecycle, metrics and events; Orleans, Hangfire and MassTransit scaffolds No persisted run, step, attempt, lease, approval or checkpoint; no Cancelled stage
Common execution entry point CONFLICTING Lifecycle, MCP, REST and DevUI paths coexist in the agent host Route every entry through one application port
Model adapters EXISTING OpenAI, Azure OpenAI, Ollama, Azure AI Inference through IChatClient No Anthropic adapter (D16)
Model profiles, routing and fallback PARTIAL Keyed registration with missing-service fallback No profile contract, policy routing, provider failover or usage receipt
Tool discovery and MCP policy PARTIAL Explicit tool registry; MCP governance options Governance opt-in; MCP unauthenticated by default; in-process tools outside governance
Context assembly PARTIAL Instructions, file skills, text search, in-memory history No budget, provenance, snapshot or ranking
Knowledge ingestion and retrieval PARTIAL Chunking, embeddings, retry, upsert, top-k search Post-search filtering, fixed 1,536 dimensions, InMemory-only wiring
Vector store CONFLICTING Qdrant connector referenced; the agent host throws for every non-InMemory provider ADR-0013 profiles exist on paper only
Knowledge graph and code graph MISSING (service) Canonical Graphify guidance; AgentKit removed Graphify on 2026-09-13 No Factory graph service
SaaS descriptor import and generation PARTIAL Five consumer repositories with descriptors and schemas; the generator specification Decided as an import format (ADR-0025); no import or recipe service exists
Memory PARTIAL In-memory chat history; conversation storage in templates No governed memory lifecycle
Tenant infrastructure EXISTING (foundation) / PARTIAL (Factory) Extensions.Saas.*; canonical ADR-0100 Factory trace, tool and retrieval contracts carry no tenant, project or run scope
Identity EXISTING (foundation) OpenIddict Authorization Server and Identity templates; canonical ADR-0101 Factory roles, delegation and resource permissions exist only as design (D18)
Execution sandbox and workspaces MISSING A file-skill script runner that throws; legacy plugin runs processes unsandboxed No per-run filesystem, compute, egress or secret boundary
Human approval MISSING Prototype toasts; design pages No approval aggregate, decision record or expiry
Evaluation and replay PARTIAL Evaluator registration; unit tests No release-gated suites, datasets or replay lineage
Observability PARTIAL OpenTelemetry, Factory meter, lifecycle events No run attribution, terminal-state coverage or export verification
AI cost and quotas PARTIAL Generic tenant metering and quota template No token or pricing capture, reservation or reconciliation
Messaging port and transports EXISTING (port) / PARTIAL (Factory use) IEventBus; Orleans outbox/inbox; MassTransit and NServiceBus adapters The Factory lifecycle publishes only through the MassTransit publisher; no transactional run state plus outbox
Factory Studio MISSING (product) / EXISTING (prototype) 41 prototypes; Blazor shell and MFE templates; AIChat mock No console, no BFF, no live state
Marketplace, compute fleet, A2A, fleet operations MISSING Design pages only Not started

The rule for moving a page from Target to Current

A service, platform or agent page keeps doc_status: target until all of the following hold, and the page then carries a Current implementation section that states them:

  1. a repository and revision are named in which the described behaviour exists;
  2. the evidence is reproducible from that revision (a build, a test run, a contract test for the transport profile, or a documented inspection with its date);
  3. the model entry's status is changed to current in the same change, so the generated catalogs and the coverage page agree with the page.

A page may be partly current: the section names which aggregates, APIs or events exist and which do not. The documentation standards carry this rule; the coverage page and the model report count the pages per status.