Skip to content

roadmap.md

Latest commit

 

History

History
65 lines (55 loc) · 2.99 KB

roadmap.md

File metadata and controls

65 lines (55 loc) · 2.99 KB
Raw

Roadmap & Future Modules

Completed

  • ✅ DONE: Modular memory architecture

  • ✅ DONE: Temporal indexer (STM/LTM)

  • ✅ DONE: FSM procedural cache

  • ✅ DONE: Symbolic key-value and graph store

  • ✅ DONE: Multimodal perception adapter

  • ✅ DONE: Vision encoder module

  • ✅ DONE: Reflexion/agent integration stubs

  • ✅ DONE: Initial LLM clients (OpenAI, Claude, Ollama)

  • ✅ DONE: TDD, benchmarks, VS Code dev config

  • ✅ DONE: Semantic cache/compression

Completed Enhancements

  • ✅ DONE: Persistent world model memory
  • ✅ DONE: Real-time agentic CLI and Web UI
  • ✅ DONE: Expanded open-source LLM connectors (Llama, DeepSeek, etc.)
  • ✅ DONE: EffortEvaluator & ConfidenceRegulator for collapse resistance metrics
  • ✅ DONE: HypothesisManager and quantized state tree for multi-path reasoning
  • ✅ DONE: Procedural backtracking and fallback logic
  • ✅ DONE: Puzzle benchmark harness for algorithmic planning tasks
  • ✅ DONE: Automatic world model transition feeding from agent streams (record_perceived_action in PerceptionSession + IntegrationLayer auto path for AgentMessage). Reduces reliance on explicit triggers for basic latest-state / predictive model maintenance (part of agent-substrate-autonomy work).

Roadmap Highlights

Universal Cognitive Layer Vision (AI Era)

  • Local, private, secure memory for every device (laptop, mobile, AR/VR, automotive, edge)
  • Contextual, multimodal, explainable memory and reasoning
  • Universal API, plugin SDK, and device/OS integration
  • Edge AI, federated learning, adaptive/personalized inference
  • Resilience, performance, fault tolerance, offline-first
  • Developer ecosystem, interoperability, open standards

Strategic Gaps & Next Steps

  • Federated learning/adaptive edge AI
  • Mobile/AR/automotive SDKs
  • Visual explainability for users
  • Plugin marketplace/ecosystem
  • Open schema/standards for context/memory
  • Unified privacy/user control dashboard

SafetyGuardrail Phases

  • ✅ Phase 1: Basic checks for FSM & DB
  • ✅ Phase 2: Policy-based rules for agents & LLM
  • ✅ Phase 3: Full rollback support + external policy config file

Next Steps

The following actions reinforce the math-driven data foundation:

  • Document All Data Models – provide schemas and diagrams for each memory structure.
  • Implement Runtime Validators – check FSM reachability and graph connectivity automatically.
  • Add Property-Based Tests – stress-test symbolic and temporal modules with proptest.
  • Pilot Statistical Monitoring – collect moving averages and standard deviation for key metrics.
  • Automate Observability Dashboards – integrate logs and metrics in the web dashboard using the new MonitoringService and Tauri UI.
  • Deploy Enhancement Advisor – surface reasoning-based suggestions for users to approve and refine.
  • Local inference via Ollama or custom backends

Post-MVP Ideas

  • Persistence for FSM backend
  • Advanced LLM plugin hosting
  • Semantic cache eviction policies
  • Multi-dashboard views for admins