Changelog

What's New?

All releases and updates from the Zensation ecosystem.

ZenBrainv0.2.1

Patch Release — Stability & Compatibility

  • Improved TypeScript strict-mode compatibility across all algorithm exports
  • Minor fixes to numeric edge cases in Hebbian weight normalization
  • Updated peer dependency ranges for broader ecosystem compatibility
  • All 276 tests passing, zero regressions
ZenAIPhase 145

Predictive Memory Architecture — 8 Neuroscience Services, 443 Tests

  • Predictive Memory Architecture (PMA): 8 neuroscience-grounded services for memory reconsolidation, stability protection, and metacognitive monitoring
  • NeuromodulatorEngine: 4-channel tonic/phasic modulation (dopamine, norepinephrine, serotonin, acetylcholine)
  • ReconsolidationEngine: prediction-error-gated memory updates with 4 modes (confirmed, selective_edit, integration, new_episode) and rollback support
  • TripleCopyMemory: fast/medium/deep copy cascade with divergent strength dynamics (Basel 2024 model) and sudden recall detection
  • StabilityProtector: NogoA/HDAC3 analog lock scoring with rigidity factors that increase with memory age
  • MetacognitiveMonitor: positivity bias detection, urgency scoring, prediction error tracking, and efficiency metrics
  • SemanticPreClusterer: centroid-based memory clustering with HNSW-indexed 768-dim embeddings
  • PriorityMap: importance-weighted memory retrieval with novelty and emotional salience scoring
  • STCRescue: synaptic tag-and-capture rescue mechanism for at-risk memories during sleep-compute
  • 10 new database tables per schema (104 tables total), 19 ablation feature flags
  • 13 new API endpoints for PMA memory and metacognition
  • All services degrade gracefully when migration tables are missing (42P01 handling)
  • 443 new tests (8 suites), total: 11,589 tests passing
ZenAIPhase 143

Agent Ecosystem Expansion — Blueprints, Marketplace, Smart Triggers

  • 8 Built-In Agents: Email Triage, Meeting Prep, Daily Digest, Research Monitor, Content Calendar, Expense Tracker, Task Prioritizer, Knowledge Curator
  • Blueprint Registry replaces 3 separate template systems — unified CRUD + import/export
  • Natural Language Agent Builder: describe an agent in natural language, Claude generates a blueprint with tool allowlist and validation
  • Cross-Learning between agents: insight extraction from successful executions (rating >= 4), EMA-based tracking
  • Smart Triggers: composite AND/OR/SEQUENCE matching, context-aware scheduling (skip/delay on busy/focus)
  • Agent Analytics: system overview, per-agent performance, optimization suggestions (underused/inefficient/over-budget)
  • A2A Discovery: remote agent card indexing, keyword-based task matching, health refresh
  • Marketplace: community browse with filters/sorting, install/publish flow, rating system
  • 6-Tab Frontend: My Agents, Create, Marketplace, Analytics, Workflows, A2A
  • 222 new tests (11 suites)
ZenAIPhase A11y

Accessibility & CSS — axe-core Coverage, ARIA Fixes, Tailwind v4

  • axe-core coverage expanded to 6 pages, all ARIA violations fixed
  • Missing shadcn/ui color variables added to Tailwind v4 @theme block
  • Font paths corrected
  • Radar chart labels no longer clipped, flywheel overlap fixed
  • Tool count display updated from 55 to 60
  • Website SEO: OG images added for all comparison pages (/vergleich/*)
ZenAIPhase DevOps

Playwright E2E, Migration Runner, Docker Hardening

  • Playwright E2E: 5 test suites, 24 cases, CI pipeline with API mocking
  • Migration Runner CLI: SHA-256 checksums, drift detection, health endpoint
  • Dockerfile HEALTHCHECK + docker-compose.dev.yml for local development
  • .env.production.example with full documentation of all variables
  • Website sync automation: GitHub Actions auto-sync on push to website/**
ZenAIPhase Multi-Tenancy

Multi-Tenancy — Organizations, Invitations, RBAC

  • Organization member management: email invitations, role assignment
  • RBAC: Owner, Admin, Member — granular permissions per resource
  • React Query hooks for all multi-tenancy operations
  • Workspace isolation: each organization has dedicated database paths
  • 38 new tests for org flows
  • @zensation integration: Hebbian/ltm-utils re-exports + algorithms/index.ts barrel
ZenAIPhase Billing

Billing & Payments — Stripe Checkout, Credits, BillingTab

  • Stripe Checkout, Customer Portal and Webhooks fully integrated
  • Credits system: usage-based billing with automatic top-up
  • BillingTab in settings: subscription management, invoice history, plan upgrade
  • PricingPage: 4 tiers (Free/Pro/Team/Enterprise) with annual toggle
  • 61 tests for all billing flows
  • SEPA direct debit for EU customers
ZenBrainv0.2.0

MemoryCoordinator + Sleep Consolidation

  • MemoryCoordinator orchestrates all 7 layers (store, recall, consolidate, decay, review)
  • Sleep Consolidation: Memory replay simulation (Stickgold & Walker 2013)
  • 95% confidence intervals for all probabilistic outputs
  • Retention curve visualization for Ebbinghaus curves
  • Tests: 153 to 276 (+80%)
  • Numeric stability: epsilon guards in Hebbian, logger parameters
  • 7 new docs: CONTRIBUTING, CoC, SECURITY, ROADMAP, FAQ, Benchmarks, API Reference
ZenBrainv0.1.0

ZenBrain Open Source Launch

  • @zensation/algorithms@0.1.0: 12 neuroscience algorithms, zero dependencies
  • @zensation/core@0.1.0: 7 memory layer implementations
  • @zensation/adapter-postgres@0.1.0: PostgreSQL + pgvector
  • @zensation/adapter-sqlite@0.1.0: SQLite zero-config
  • Apache 2.0 license — free forever
  • GitHub repository: github.com/zensation-ai/zenbrain
  • Zenodo DOI: 10.5281/zenodo.19353663 (technical paper)
  • HuggingFace Model Card: alexanderbering/zenbrain
  • 153 tests, TypeScript, zero dependencies
ZenAIPhase 142

ZenAI V4 — Multi-LLM, AG-UI, GraphRAG Events, HyperAgents

  • Multi-LLM Routing in chat stream: model-orchestrator with Redis LLM cache, 6 providers active
  • AG-UI Protocol: 6 cognitive event emitters (hypothesis, prediction, review, pipeline, approval)
  • Adaptive Thinking frontend: 4 display modes for thinking_tier SSE (hidden/collapsible/visible/progress)
  • GraphRAG event recording in tool-execution, hybrid-retriever and chat + daily BullMQ pruning worker
  • HyperAgents: Self-Improvement → meta-improver bridge + weekly observation job
  • CLI renamed to @zensation/cli, npm-ready with 108 tests
  • Dual-emit SSE + AG-UI for all cognitive events
ZenAIPhase 125–140

Cognitive Architecture — 16 Phases, 30+ Services, 1100+ Tests

  • Hebbian Knowledge Graph: co-activation, decay, normalization (Phase 125)
  • FSRS Spaced Repetition + Bayesian confidence propagation (Phase 125)
  • Pinned Core Memory Blocks (Letta pattern), Cross-Context Entity Merging (Phase 126)
  • Global Workspace Theory: competitive context assembly, 8 specialist modules (Phase 127)
  • Chain-of-Thought persistence with pgvector similarity, Multi-Hop inference (Phase 128)
  • Persistent Agent Loops: checkpoints, pause/resume, context isolation (Phase 129)
  • Tool Composition Engine, Dynamic Team Builder with 5 specialists (Phase 130)
  • Document export: PPTX, XLSX, PDF, DOCX — directly from chat (Phase 131)
  • Curiosity Engine: knowledge gap detection, information gain scoring (Phase 133)
  • Prediction Engine: temporal pattern recognition, prediction error learning (Phase 134)
  • Confidence calibration, confusion detection, capability profiling (Phase 135–136)
  • Feedback Bus + Adaptive Behavior Engine + Style Learner (Phase 137–138)
  • Cross-Pillar Pipeline + Self-Improvement Engine (Phase 139–140)
  • 1100+ new tests across all 16 phases
ZenAIPhase 132

CLI Agent — @zensation/cli with 108 Tests

  • Agent Loop: Claude Code pattern — single-threaded loop with tool execution and max-iterations guard
  • Filesystem Tools: read_file, write_file, edit_file, list_files, search_content, run_command
  • Backend Bridge: API client for Memory (remember/recall), Web Search, Idea Search, Core Memory
  • Terminal UI: Chalk-based output with Markdown rendering and spinner
  • Context Management: .zenai/ directory, session persistence (JSON), project detection (11 types)
  • Hybrid architecture: local file tools + backend API for Memory/KG/RAG
  • Prepared as @zensation/cli npm package
ZenAIPhase 141

Cognitive Architecture Frontend + Mistral Integration

  • CognitiveDashboard redesign with 5 sub-tabs
  • 23 new API endpoints (Curiosity, Predictions, Feedback, FSRS, Self-Improvement)
  • 14 React Query hooks for all cognitive services
  • Gamified FSRS review queue with 5-grade rating
  • Chat feedback buttons (thumbs up/down)
  • JWT auth fix: apiKeyAuth correctly delegates JWT tokens
  • Mistral AI integration as cloud fallback
  • ESLint 596 to 0 warnings (130 files)
ZenAIPhase 119–124

Quality & Test Coverage Sprint — 1100+ New Tests, File Decomposition

  • tool-use.ts decomposed from 1689 to 215 LOC (Phase 119)
  • SystemAdminPage split into 6 separate files (Phase 120)
  • GeneralChat, ProceduralMemory, AgentTeams components extracted (Phase 121)
  • ideas-handlers + general-chat-handlers extracted from routes (Phase 122)
  • 358 new tests across 30 files in a single phase (Phase 123)
  • 351 new tests + all 12 Dependabot security vulnerabilities fixed (Phase 124)
  • Route test coverage increased from 38% to 98%
  • Backend 7720 + Frontend 1400 + CLI 108 = 9228 tests, 0 failures
ZenAIPhase 100

Deep Excellence — Self-Editing Memory, Parallel Agents, Chat UX

  • Self-Editing Memory: memory_replace, memory_abstract, memory_search_and_link
  • Contextual Retrieval (Anthropic method, +67% retrieval accuracy)
  • CRAG Quality Gate: automatic query reformulation on low confidence
  • LLM-based memory consolidation with token budget management
  • Parallel Agent Execution + Persistent Shared Memory
  • Dynamic Agent Teams: 5 specialist roles, semantic tool search
  • Chat Edit & Regenerate: tree branching for conversation history
  • Persistent tool disclosure + expandable thinking display
  • Glass variants, inline error recovery, confidence indicators in design system
ZenAIPhase 97–99

Quality Excellence — 59 Fixes, Security, RAG, State Machine

  • SQL injection fix, encryption enforcement, auth rate limits (Phase 97)
  • 60 composite indexes, FK constraints, pool monitoring (Phase 97)
  • Contextual Retrieval, Tool Search Tool, Dynamic RAG Weights, Self-RAG Critique (Phase 99)
  • Embedding Drift Detection with BullMQ worker (Phase 99)
  • Chat State Machine (chatReducer.ts): centralized state automaton (Phase 99)
  • Request timeout, tool limits, error sanitization (Phase 99)
  • React Query migration: 4 to 8 modules (Phase 97)
  • ARIA live regions, keyboard navigation, page skeletons (Phase 97)