-
-
Notifications
You must be signed in to change notification settings - Fork 2
Roadmap
Bellok edited this page Mar 28, 2026
·
8 revisions
Back to Home
This page tracks the milestones from the current research preview toward the 1.0.0 production release.
The project is in active pre-1.0 development. All 0.x releases should be interpreted as:
- Ready for research, controlled pilots, and single-agent workflows.
- Not ready for production-scale, multi-tenant, or high-criticality deployments.
This position is intentional and consistent with semantic versioning conventions for 0.x software.
Focus: Operational robustness and observability
- Structured logs with consistent schema
- Actionable error taxonomy (categories, codes, recovery hints)
- Write serialization strategy (prevent concurrent-write corruption)
- Crash-recovery strategy (WAL or equivalent)
- Documented scale and performance envelope (memory count, retrieval latency baselines)
Focus: Multi-agent foundation and lifecycle semantics
- Lifecycle hooks for memory events (on-write, on-retrieve, on-delete)
- Memory-event semantics (publishable, subscribable events)
- Orchestration-aware retrieval strategies
- Goal-aware retrieval and ranking
- Conflict-aware write and merge patterns
- Council / Kanban multi-agent reference implementations
- Reference orchestration patterns for parallel agent execution
Focus: Production readiness and API stability
- Durability and consistency guarantees (documented and tested)
- Stable API contract with migration guidance for breaking changes
- Validated operational runbooks
- Observability and operations telemetry (metrics, tracing, alerting guidance)
- Orchestration-native control surfaces
The long-term direction of this project is supporting council-style and Kanban-like agent orchestration, where multiple CLI agents coordinate through shared memory with explicit lifecycle semantics.
This is designed for:
- Parallel agent execution — multiple agents reading and writing memory concurrently.
- Role-differentiated agents — agents assigned to specific cognitive districts.
- Shared context without coupling — agents communicate through memory events rather than direct calls.
- Neurodivergent-native workflows — non-linear, associative task management rather than strict sequential pipelines.
See also: Release-Notes · Architecture · White-Paper