Extension Roadmap¶
Target Architecture — Final-State Design
This page describes how the Knowledge Platform extends beyond its final-state baseline. The 22 services, 20 workers, 32 aggregates, and 14 APIs documented elsewhere in this section are the stable core; the items here are additive, designed to plug into that core without breaking its contracts.
The Knowledge Platform is built to grow. New knowledge sources, retrieval strategies, governance regimes, and agent capabilities can be added as new bounded contexts, services, and workers that integrate through the canonical event envelope and the existing aggregates — never by modifying the core contracts. This keeps the platform's promises of traceability, reusability, autonomy, governance, observability, and multi-tenant scale intact as it evolves.
Extension Principles¶
- Additive, not invasive — new capabilities arrive as new services/workers subscribing to existing events, not as changes to core aggregates or APIs.
- Contract-first — every extension declares its events (
NounVerbPastTense), commands (VerbNoun), and API shapes up front, versioned per the naming conventions. - Governed by default — any new knowledge source flows through classification, access policy, and quality before it can enter a context package.
- Backend-abstracted — new stores or models plug in behind existing ports (as Qdrant ↔ Azure AI Search does today), so callers are unaffected.
- Traceable & observable — extensions carry the cross-cutting metadata fields and emit OTEL telemetry from day one.
- MCP-exposed — agent-facing extensions are surfaced as tools via
ConnectSoft.Extensions.ModelContextProtocolso Agent Mesh agents (Microsoft Agent Framework) can use them.
Future Microservices¶
| Candidate Service | Purpose | Integrates via |
|---|---|---|
KnowledgeGraphReasonerService |
Multi-hop graph reasoning and inferred-edge suggestion over the knowledge graph | Graph projections, INFORMED_BY edges |
CrossTenantPatternService |
Govern sanctioned cross-tenant pattern sharing with anonymisation | KnowledgePatternDiscovered, access policy |
KnowledgeSummarizationService |
Produce tiered summaries (executive, technical) of large knowledge clusters | MemoryRecord, vector memory |
OntologyManagementService |
Manage and version the graph ontology (node/edge types) as governed config | Knowledge graph |
SemanticDiffService |
Compute semantic (not textual) diffs between artifact versions | Artifact memory, embeddings |
KnowledgeFreshnessService |
Proactive freshness scoring and re-validation scheduling | Quality, stale-memory signals |
ExternalKnowledgeConnectorService |
Govern ingestion of approved external sources (standards, vendor docs) | Ingestion, classification |
Future Workers¶
| Candidate Worker | Trigger | Purpose |
|---|---|---|
GraphInferenceWorker |
GraphProjectionUpdated |
Suggest inferred edges (likely dependencies, reuse) for human confirmation |
SummaryRefreshWorker |
Scheduled / MemoryRecordCreated |
Keep cluster summaries current |
KnowledgeDriftWorker |
Scheduled | Detect when memory diverges from current code/runtime reality |
CostOptimizationWorker |
Scheduled | Tier/evict cold vectors and blobs by usage signals |
EmbeddingModelEvalWorker |
EmbeddingModelUpgraded |
A/B evaluate a new embedding model before full refresh |
CrossProjectReuseWorker |
KnowledgePatternDiscovered |
Surface reuse candidates into new projects |
Future APIs¶
| Candidate Endpoint | Purpose |
|---|---|
POST /knowledge/graph/reason |
Multi-hop reasoning / path explanation over the graph |
POST /knowledge/summarize |
On-demand tiered summary of a knowledge set |
POST /knowledge/diff/semantic |
Semantic diff between two artifact versions |
GET /knowledge/freshness/{ref} |
Freshness and validity score for a memory item |
POST /knowledge/context/explain |
Explain why each source was selected for a context package |
POST /knowledge/reuse/suggest |
Suggest reusable patterns/blueprints for a new project intent |
All new endpoints honour the existing authorization, versioning, and governance model documented in APIs.
Marketplace Opportunities¶
The platform's reusable knowledge is a natural feeder for the Marketplace:
- Pattern listings — high-confidence
KnowledgePatterns (Public-classified) published as marketplace assets for cross-organisation reuse. - Curated context packs — domain-specific, governed bundles of decisions, patterns, and blueprints for common verticals.
- Embedding/ontology packs — pre-built ontologies and domain embeddings tenants can adopt to bootstrap a project.
- Quality rule packs — shareable
QualityRulesets encoding organisational or industry standards.
Agent Opportunities¶
New agent capabilities the platform can ground (via MCP tools and the Agent Mesh):
- Knowledge Curator agent — reviews flagged stale/conflicting knowledge and proposes resolutions.
- Reuse Scout agent — proactively scans new project intents and recommends existing patterns/blueprints to reuse.
- Decision Historian agent — answers "why did we decide X?" by walking the decision graph and citing
DecisionRecords. - Incident Analyst agent — correlates new incidents with historical
IncidentMemoryand the producing context to suggest fixes. - Ontology Steward agent — proposes ontology refinements as the factory's knowledge shape evolves.
These agents consume governed context like any other, so their reasoning remains traceable, auditable, and grounded.
Roadmap Themes¶
flowchart LR
Core["Final-State Core<br/>(22 svc, 20 wkr, 32 agg, 14 API)"] --> Reason["Graph reasoning & inference"]
Core --> Reuse["Cross-tenant reuse & marketplace"]
Core --> Quality["Proactive freshness & drift"]
Core --> Explain["Context explainability"]
Reason --> Agents["New grounded agents"]
Reuse --> Agents
Quality --> Agents
Explain --> Agents