Status: Living Document
Project: CKS Runtime
Current Target: Version 2.0 buildout (Runtime 1.0 was reached and surpassed; actual latest release is v1.33.0)
Roadmap Version: 1.2
This roadmap defines the long-term evolution of CKS Runtime from an operational execution library into the canonical operational platform for Canonical Knowledge Structures.
The roadmap serves as the primary strategic planning document for Runtime development.
Phases 1 through 8 below (0.2.x through 1.0.0) are complete — the
project has shipped well past Runtime 1.0. Verified against
CHANGELOG.md:
- Distributed Runtime — Replication: peer discovery
(
PeerDiscovery,HTTPPeerDiscovery), weighted peer selection with backoff (PeerScheduler), and background anti-entropy cycles (GossipService) — done (ADR-008). - Reasoning-conflict detection:
InferenceStalenessSweeper, running on a configurable interval, escalatesInferenceConflictDetectedfindings — done (ADR-009). The consumption side (draining this queue into an autonomous agent) is intentionally out of scope forcks-runtimeand is tracked as the Critic Agent milestone incks-mcp's roadmap — see that repository'sROADMAP.md. - Storage: PostgreSQL backend (async storage ABC, pgvector support) alongside SQLite — done.
- Execution Engine (partial):
ValidateOperation,EvolveOperation,SerializeOperation,ExplainOperationimplemented viaCoreBridge— in progress, see Version 2.0 below for the remaining scope (dependency resolution, parallel execution, retry/compensation).
The granular sub-items under "Version 1.x — Production Runtime"
below (1.1–1.9) have not all been individually re-verified
against CHANGELOG.md in this pass — Replication (1.4) is
confirmed done; the rest of that section should get a dedicated
audit pass rather than being assumed complete just because the
project is past v1.0.
CHARTER
↓
ARCHITECTURE
↓
ADR
↓
SPECIFICATIONS
↓
ROADMAP
↓
IMPLEMENTATION
| Document | Purpose |
|---|---|
| CHARTER | Vision & philosophy |
| ARCH | Runtime architecture |
| ADR | Architectural decisions |
| SPEC | Runtime standards |
| ROADMAP | Evolution strategy |
CKS Runtime provides the canonical operational platform for Canonical Knowledge Structures.
Runtime owns execution.
CKS Core owns semantics.
Runtime coordinates:
- Sessions
- Transactions
- Execution
- Persistence
- Versioning
- Diagnostics
- Events
- Explainability
- Storage
- Adapters
- Plugins
Runtime is intentionally not a semantic engine.
Runtime Library
↓
Execution Runtime
↓
Operational Runtime
↓
Production Runtime
↓
Runtime Platform
↓
Cloud Runtime
↓
Autonomous Runtime
↓
Semantic Operating Environment
| Version | Runtime maturity |
|---|---|
| 0.2 | Foundation |
| 0.3 | Execution Runtime |
| 0.4 | Event Runtime |
| 0.5 | Adapter Runtime |
| 0.6 | Plugin Runtime |
| 0.8 | Production Runtime |
| 1.0 | Stable Platform |
| 2.0 | Runtime Platform |
| 3.x | Cloud Runtime |
| 4.x | Autonomous Runtime |
| 5.x | Semantic Operating Environment |
The following principles are architectural invariants.
These principles should remain valid across all future major releases.
Runtime coordinates execution.
Runtime never defines semantic meaning.
Validation.
Evolution.
Serialization.
Explanation.
Reasoning.
Semantic correctness.
These responsibilities always belong to CKS Core.
Communication with semantic engines occurs exclusively through
- CoreInterface
- CoreBridge
Runtime never imports implementation-specific Core objects.
Every replaceable subsystem depends only on abstractions.
Examples
- Storage
- Core
- Adapter
- Operation
- Plugin
The same Runtime state and the same sequence of operations must always produce the same operational result.
Whenever practical Runtime public models remain immutable.
Examples
- RuntimeVersion
- RuntimeValidationResult
- RuntimeEvent
- Diagnostic
Runtime never depends on
- MCP
- HTTP
- CLI
- WebSocket
- gRPC
Adapters depend on Runtime.
Never the opposite.
Runtime operates independently of storage implementations.
Reference implementation:
- InMemoryStorage
Production implementations:
- Plugins
Every significant Runtime action is observable.
- Sessions
- Transactions
- Operations
- Versions
- Events
- Diagnostics
Every subsystem should be independently testable.
Dependency injection is preferred over global state.
Runtime intentionally is not
- a semantic engine
- an ontology framework
- a reasoning engine
- a graph database
- an inference engine
- a knowledge validator
- a query language
These responsibilities belong to CKS Core or external plugins.
Foundation
↓
Execution
↓
Events
↓
Adapters
↓
Plugins
↓
Production
↓
Specification Freeze
↓
Runtime 1.0
Complete the operational architecture.
- Runtime façade
- CoreBridge
- CoreInterface
- RuntimeValidationResult
- Sessions
- Transactions
- Versions
- Storage abstraction
- Diagnostics
- Execution Pipeline
- Runtime contains no semantic logic
- Runtime depends only on CoreInterface
- CoreBridge is the exclusive semantic gateway
- Unit tests passing
Introduce canonical Runtime execution.
- Dispatcher
- OperationRegistry
- OperationExecutor
- ExecutionContext
- ExecutionResult
- Built-in Runtime Operations
Operations
- Validate
- Serialize
- Explain
- Evolve
- Query
- Transform
Execution always follows
Dispatcher
↓
Registry
↓
Executor
↓
CoreBridge
↓
ExecutionResult
Complete Runtime lifecycle events.
- Session lifecycle
- Transaction lifecycle
- Version lifecycle
- Operation lifecycle
- Diagnostic lifecycle
- EventBus
- Subscriber API
- Event history
Every Runtime lifecycle transition emits RuntimeEvents.
Formalize Runtime integrations.
- Adapter API
- MCP Adapter
- HTTP Adapter
- CLI Adapter
- Python Adapter
Future
- gRPC
- WebSocket
Runtime remains completely transport independent.
Completely decouple Runtime from implementations.
- Plugin Manager
- Plugin Discovery
- Plugin Metadata
- Plugin Lifecycle
Plugin Interfaces
- Core
- Storage
- Adapter
- Operation
- Explainability
Official plugins
- cks-runtime-plugin-cks-core
Future
- RDF
- OWL
- Neo4j
- NetworkX
- OpenAI
Runtime depends only on abstract interfaces.
Production-grade operational platform.
- Recovery API
- Crash recovery
- Session recovery
- Transaction recovery
- Pipeline recovery
- Snapshots
- SQLite
- PostgreSQL
- Redis
- File
- Object Storage
- Nested
- Optimistic locking
- Conflict detection
- Compensation
- Diff
- Checkout
- Restore
- Branching
- Merge
- Structured codes
- Categories
- Correlation IDs
- Trace IDs
- Logging
- Metrics
- Tracing
- Profiling
- Telemetry
Runtime is suitable for production deployments.
Freeze public architecture.
Review
- Public API
- Plugin API
- Adapter API
- Storage API
- Documentation
- Conformance
No unresolved architectural questions remain.
First stable Runtime platform.
Requirements
- Stable Runtime API
- Stable Plugin API
- Stable Adapter API
- Stable Storage API
- Complete documentation
-
95% automated tests
- Compatibility guarantees
Breaking changes require a major version.
Recovery
Snapshots
Crash-safe execution
Lease management
Metrics
Tracing
Execution timeline
Profiling
Production storage
Migration framework
Backup API
Distributed Sessions
Distributed Transactions
Replication
Leader election
Plugin SDK
Registry
Dynamic loading
Compatibility management
Pipeline optimisation
Parallel execution
Caching
Memory optimisation
Authentication
Authorization
Capability model
Sandbox
Audit trail
Docker distribution was considered and intentionally descoped — not needed, and it kept causing confusion about what's actually required to run the project. Not planned for
cks-runtimeor the wider ecosystem.
Kubernetes
Helm
Runtime Service
LTS
Operational hardening
API stability
Runtime becomes self-describing.
Status of the sub-goals below, verified against CHANGELOG.md
(August 2026):
| Sub-goal | Status |
|---|---|
| Runtime Graph | Not started |
| Runtime Introspection | Not started |
| Execution Engine | In progress — ValidateOperation/EvolveOperation/SerializeOperation/ExplainOperation shipped via CoreBridge; dependency resolution, parallel execution, retry, and compensation still open |
| Execution Plans | Not started |
| Runtime Scheduler | Not started as a general-purpose feature (the reasoning-specific InferenceStalenessSweeper does periodic background execution, but that's narrower than this goal) |
| Reactive Runtime | Partial foundation — the Event Bus is extensively used, but "everything event-driven" is not yet the case |
| Runtime DSL | Not started |
| Distributed Runtime | In progress — Replication shipped (gossip, ADR-008); Runtime Cluster, Shared Storage, and a general Distributed Event Bus still open |
| Observability Platform | Not started as a unified platform (basic metrics/event logs exist in cks-mcp via get_metrics, but no Timeline/Replay/Heatmap/Explorer/Profiler) |
| Runtime Studio | Not started |
Sessions
Transactions
Versions
Operations
Diagnostics
Events
become one operational graph.
Supports
- Dependency graph
- Execution graph
- Version graph
- Diagnostic graph
- Event graph
Runtime explains itself.
Examples
- Why did this transaction fail?
- Which operation created this version?
- Which diagnostics were produced?
- Dependency resolution
- Parallel execution
- Retry
- Compensation
- Pipeline optimisation
Declarative execution pipelines.
- Scheduled execution
- Deferred execution
- Background execution
- Periodic execution
Everything becomes event-driven.
Declarative Runtime workflows.
- Runtime Cluster
- Shared Storage
- Distributed Event Bus
- Replication
- Timeline
- Replay
- Heatmap
- Graph Explorer
- Profiler
Visual Runtime environment.
Manage
- Sessions
- Transactions
- Versions
- Operations
- Events
- Diagnostics
- Execution Graphs
- Multi-tenancy
- Horizontal scaling
- Federation
- Cloud-native deployment
- Managed Runtime
- Adaptive scheduling
- Self-healing
- Intelligent optimisation
- Policy engine
- Runtime recommendations
Runtime becomes the operational kernel of the complete CKS ecosystem.
Before Runtime 1.0
Architecture may evolve when necessary.
After Runtime 1.0
- Patch releases fix defects only.
- Minor releases add backward-compatible functionality.
- Major releases introduce architectural changes.
This roadmap evolves together with
- Runtime Charter
- Runtime Architecture
- ADRs
- Runtime Specifications
- Conformance Suite
Every completed phase should update the corresponding architecture and specification documents.
CKS Runtime favours architectural stability over implementation complexity.
Every capability must preserve
- Runtime/Core separation
- Operational determinism
- Storage independence
- Transport independence
- Adapter independence
- Plugin independence
- Semantic authority of CKS Core
CKS Runtime shall never become a second semantic engine.
CKS Core remains the single source of semantic truth.