Changelog

/reviews community page — new page for users to leave reviews, showcase projects, and ask questions. Form submits via /contact API. Three review types: review, project showcase, question.

/benchmark page — full latency and throughput benchmark page. Real p50/p95/p99 numbers for /recall, /inject-context, /remember. Infrastructure details, confidence scoring explainer, comparison table.

/compare-memstate — new head-to-head comparison page added to the compare hub. Coverage: integration, compliance, transparency, pricing, migration guide.

Docs: 7 new agent setup guides — Kilo Code, OpenCode, ChatGPT, Grok SDK, Claude Agent SDK, Google ADK, Vercel AI SDK. All with Automatic/Manual tabs and npx @kronvex/setup one-liner.

Docs: Use Cases SPA page — merged /use-cases into docs. 5 main cases with code examples. /use-cases redirects to /docs#use-cases.

Homepage: live memory stream terminal — "Every agent. One memory." section redesigned as an animated live feed showing agent writes/reads in real-time. Replaces static diagram.

Homepage: conflict detection section — visual terminal showing confidence-scored memory deduplication. Formula display with color-coded weights.

Demo: UPLOAD → INGEST → QUERY — step labels and copy updated to match actual flow. Auto-fire on page load removed.

@kronvex/mcp v1.1.0 + @kronvex/setup v1.1.0 — KRONVEX_AGENT_ID env fallback, agent_id optional in all tool schemas, setup auto-creates "default" agent.

#how section background — fixed CSS specificity issue where background: transparent !important overrode the intended var(--bg2) background on the "How it works" section.

Integrations page merged into docs — /integrations now redirects to /docs#integrations. All nav links updated.

CLI kronvex (B3) — pip install kronvex installs a full terminal client. Commands: agents list/create/delete · remember · recall · memories list. Env-var auth via KRONVEX_API_KEY. JSON output by default, --pretty for human-readable.

Extended webhook events (B2) — 4 new events: memory.recalled · memory.deleted · memory.expired · memory.consolidated. Retry logic: 3 attempts with 2s/8s exponential backoff. All dispatches non-blocking via BackgroundTasks. Fresh DB session per dispatch — no session leak.

n8n node update (B1a) — New health and consolidate operations. recall gains optional context_messages JSON field for A4 re-ranking.

Dify plugin (B1b) — Native Dify plugin at sdk/dify/ with remember, recall, and inject_context tools. YAML manifests, Python tool classes, credential validation via /health.

MCP server (B1c) — FastMCP server at sdk/mcp-server/ exposing 4 tools: remember · recall · inject_context · list_agents. Plug directly into Claude Desktop or any MCP-compatible host via env vars.

Memory auto-consolidation (A1) — background job clusters semantically similar memories (cosine ≥ 0.82) and merges them into a single meta-memory via GPT-4o-mini. Source memories soft-deleted with audit trail. Dashboard shows ⬡ META badge on consolidated memories.

Entity extraction + knowledge graph (A2) — every remember call now triggers an async GPT-4o-mini pass extracting named entities (person, org, preference, fact, procedure) and their relations into two new tables: entities and entity_relations. Recall runs vector search + graph traversal, results merged and re-ranked by confidence.

Type-aware memory decay (A3) — five memory types with distinct sigmoid decay curves: fact (180-day), preference (60d), procedure (90d), context (3d), episodic (14d). Auto-classification via GPT-4o-mini when memory_type is omitted.

Contextual re-ranking (A4) — recall and inject-context accept a new optional context_messages[] field (last N conversation turns). GPT-4o-mini re-ranks results using full conversation context for significantly better precision in multi-turn dialogues. Fully backwards-compatible.

Memory health score (A5) — new endpoint GET /agents/{id}/health returns four metrics: coverage (semantic breadth), freshness (weighted age), coherence (contradiction detection), utilization (recalled in last 30 days). Surfaced in the dashboard with actionable text recommendations.

Dashboard: health widget & META badges — health score visible on each agent card. Consolidated meta-memories display ⬡ META badge in the memory explorer. New type colours for fact, preference, and context memory types.

Migration 005 — adds is_meta, consolidation_count, consolidated_from (JSONB), user_id columns to memories. Creates entities + entity_relations tables with full indexes. Applied on Supabase.

German landing page /de — new EU market page targeting German-speaking developers. Full translation, localised copy and CTAs.

French i18n — complet — full French translation of footer, demo modal validation messages, auth modal and /fr/ page links. All validation errors now display in the correct locale.

Architecture diagram animations — hover lift+glow effect on each block; SVG traveling dot arrows visualise real-time data flow between components.

Works-with emoji icons — replaced coloured dots with emoji icons on all 11 integration pills in the works-with bar (LangChain, CrewAI, AutoGen, n8n, Python SDK, Node.js SDK…).

Weaviate & LlamaIndex compare pages — two new competitor comparison pages (/compare-weaviate, /compare-llamaindex) covering feature matrix, pricing and GDPR compliance side-by-side.

SEO — schema & internal links — JSON-LD schema enriched (breadcrumbs, FAQ, SoftwareApplication), meta description fixes, new internal cross-links between DPA, industry pages and blog articles. /faq added to sitemap with industry cross-links.

EU homepage refresh — richer works-with bar, updated architecture section copy, DPA page linked from footer, FAQ schema added to homepage.

Nav auth flash & i18n shift — auth state and locale are now applied synchronously before first paint, eliminating the flickering login/logout button and language shift visible on hard reload.

Search button FOUC — layout shift caused by the search button slide-in animation on DOMContentLoaded is eliminated; button is now stable from first render.

Compare page canvas — particle canvas restricted to the hero section only; blurred background added behind sub-text for readability on all compare pages.

Pricing consistency — Scale €1 499 — Scale plan price corrected from €999 to €1 499 across all compare pages, pricing tables, blog posts and CGV.

Blog pricing pass — all blog articles and the launch post updated to reflect current plan lineup (Dev €19 entry, revised Starter/Pro/Growth/Scale values).

scrollbar-gutter: stable — added globally to prevent layout shift when modal overlays add/remove the scrollbar.

.html duplicate content — Cloudflare Worker now issues 301 redirects from /page.html to /page to prevent duplicate-URL indexing by search engines.

Section dividers — visual dividers added between the API / Architecture sections and between the verticals / "Why Kronvex" sections on the homepage for better reading rhythm.

Dev plan (€19/mo) — new entry-level paid tier with 3 agents, 5 000 memories and daily rate limits (500 recall / 1 000 remember calls). Fills the gap between the free Demo and Starter plans.

Pricing restructure — all paid plans updated: Starter €49 / 5 agents / 15k, Pro €249 / 10 agents / 75k, Growth €599 / 30 agents / 300k, Scale €1 499 / unlimited. CGV updated accordingly.

GDPR Art. 17 erasure endpoint — DELETE /agents/{id}/memories/user/{user_id} bulk-deletes all memories for a given end-user, enabling right-to-erasure compliance workflows for Pro+ customers.

Audit trail endpoint — GET /agents/{id}/audit returns a timestamped log of all API operations on an agent. Available on Pro and above.

Cosine deduplication — memories with cosine similarity ≥ 0.95 are deduplicated at write time. MemoryResponse now includes a deduplicated: bool flag.

Site-wide search bar — keyboard-accessible overlay (/ or Ctrl+K) indexes 32 pages with arrow-key navigation and instant filtering. Accessible from the navbar on all pages.

Cookie consent banner — GDPR-compliant opt-in banner with Accept / Decline, localStorage persistence (kv_cookie_consent), and animated dismiss. Loads after explicit user interaction only.

Vertical landing pages — /healthcare and /fintech industry-specific pages highlighting HIPAA/GDPR compliance, use cases, and tailored CTAs.

Umami analytics — replaced Plausible with self-hosted Umami across all 105+ HTML files. No personal data, cookie-free, GDPR-compliant out of the box.

Hero layout fix — golden left-border trait on the eyebrow badge was broken by display:inline-block. Replaced with width:fit-content to restore block formatting context while constraining background width.

Starter plan restored — Starter card was missing from the index.html pricing grid after the pricing restructure. Restored with correct values; grid updated to 5 columns; mobile slider synced.

Enterprise FAQ accordion — toggleFaq() was referenced in onclick handlers but never defined. Function added; accordion now opens/closes correctly.

llms.txt updated — current pricing, new pages (/healthcare, /fintech), and new blog articles added for AI crawler discoverability.

New compare pages — dedicated comparison pages for Zep and MemGPT, complementing the existing Mem0 and Pinecone comparisons. Each page details features, pricing, and GDPR compliance side-by-side.

Use Cases in navbar — /use-cases link added to the kn-nav desktop and mobile menu on all secondary pages (blog, pricing, docs, enterprise, roadmap, changelog, compare, use-cases).

Greek mythology background on Use Cases — page now features a subtle meander pattern, Olympian gold radial glows, and temple column lines for a richer visual identity matching the Kronvex mythological theme.

Enterprise & Compare layout — improved section spacing, hero alignment, and card grid responsiveness on /enterprise and /compare pages.

Navbar index — updated nav links on index.html to match secondary pages; consistent ordering across all pages.

Misc fixes — blog, roadmap, pricing, and docs: i18n corrections, toggle fixes, typos, and footer Contact link corrected from mailto: to /contact.

Login cache fix — Cache-Control: no-cache, must-revalidate header added to /login via Cloudflare Worker to prevent stale page serving after deployments.

Sitemap updated — /compare-zep, /compare-memgpt, /compare-mem0, /compare-pinecone, /enterprise, and /changelog added. All lastmod dates updated to 2026-03-25.

Monthly memory cycle — new cycle_memories_used counter on api_keys. Tracks memories created per billing cycle, decrements on delete. Resets to 0 on invoice.payment_succeeded Stripe webhook. Dashboard overview now shows "MEMORIES THIS CYCLE" with active count below.

Account deletion — DELETE /auth/delete-account endpoint cancels Stripe subscription, deletes all agents & memories, deactivates API key, removes Supabase user. Dashboard calls backend before signing out.

LOG OUT button in desktop nav — logout button added to the top-right desktop navbar on all secondary pages. Visible when logged in, hidden when not.

Custom cursor restored — dot cursor (grey → gold on hover) re-enabled on all secondary pages using .kv-cursor-ready *{cursor:none} — same approach as index.html. Native cursor shows on touch/coarse devices.

Light mode text readability — inject_nav now overrides h1–h6, p, li colors with !important in light mode, fixing white-on-white text caused by page stylesheet dark palette.

Native cursor on secondary pages — removed stale cursor:none!important from 404, legal, status, docs pages.

Supabase Spend Cap confirmed — Spend Cap active on Supabase Pro plan. Updated costs.html checklist.

OpenAI hard cap €50/mois — spending limit configured on platform.openai.com. Updated costs.html.

Memories tab count bug — _totalMemoriesCount was reset to 0 on each agent iteration. Fixed accumulation across all agents.

Plan detection — GET /auth/my-key now always returns the paid plan key when both a demo and paid key exist for the same user. Previous logic could return the demo key if it was sbuid-tagged first.

Key rotation — POST /auth/rotate-key now correctly sets rotated=true and rotated_at timestamp on the old key.

Deleted memories counter — new deleted_memories_count column on api_keys. Incremented on every delete (single, bulk, all, agent cascade). Displayed in dashboard overview.

Loading skeletons — animated shimmer skeleton rows in the memories list while data loads. Replaces plain "Loading…" text.

Mobile layout — nuclear fix — injected *{max-width:100vw;box-sizing:border-box} + knEnforceMobile() JS across all 54 secondary pages. Grids collapse to 1 col, flex rows stack, overflowing elements clamped.

Custom cursor removed — the dot-cursor overlay is no longer injected on secondary pages.

Favicon — favicon.svg updated to current 8-leaf crown logo. og:image added to LangGraph and MCP blog articles.

Memory TTL / expiration — pass ttl_days on POST /remember. Memories expire automatically; pinned: true overrides expiry.

Webhooks — three new events: memory.stored, quota.warning, quota.reached. Configure endpoint URL in dashboard → Settings.

Bulk import — POST /agents/{id}/memories/bulk-import ingests up to 100 memories in a single request with per-item error reporting.

API key rotation — POST /auth/rotate-key issues a new key while keeping the old one valid for 24 hours. Zero-downtime rollover for production agents.

Sentry error monitoring — exceptions automatically captured and grouped. Structured JSON logs on every API call (memory.stored, memory.recalled).

Dashboard — memory search clear button — one-click reset of the search field in the memory explorer.

Dashboard — weekly sparkline — 14-day memory storage trend chart on the overview page, powered by the new GET /api/v1/stats/weekly endpoint.

Dashboard — CSV export — download all memories for an agent as a CSV file directly from the memories page.

Rate limit headers — X-RateLimit-Limit and X-RateLimit-Window returned on every response so clients can self-throttle.

Universal navbar — unified #kn-nav on all 32 secondary pages. Identical height (60px), sticky, SVG crown logo, same Dashboard CTA colour everywhere.

Hamburger menu — slide-out mobile drawer on every page (≤768px). Previously only on the pricing page.

Blog — 4 new articles — LangGraph persistent memory, AutoGen memory, Zep alternative, MCP memory.

GDPR compliance article — promoted to main blog list. Covers EU data residency, right-to-erasure, and GDPR-native memory storage.

Dedicated /pricing page — full plan comparison table, FAQ accordion, CTA to dashboard.

Clean URL rewrites — all blog and marketing pages now served without .html extension via Cloudflare Worker.

Canonical & og:url — removed .html suffix from meta tags on 10 pages.

Recall N+1 fix — replaced per-memory UPDATE loop with a single bulk UPDATE … WHERE id IN (…).

Stripe Customer Portal — new POST /billing/portal endpoint. Users can self-manage billing, update card, cancel.

Onboarding emails — restart-safe — switched from asyncio.sleep to Resend scheduled_at. Emails survive Railway restarts.

Homepage FAQ accordion — pricing FAQ section on index replaces separate pricing page link.

Light/dark theme + FR/EN lang — localStorage-persisted, propagated across all pages via universal nav JS.

Python SDK published to PyPI — pip install kronvex. Full async support, typed.

Node.js SDK published to npm — npm install kronvex. ESM + CJS, TypeScript native.

Analytics page in dashboard — top recalled memories, type breakdown, session heatmap, growth chart (14 days).

Memories page — browse, filter, search, sort, and bulk-delete memories directly from the dashboard.

Webhook alerts — configure a URL + threshold, get notified when memory usage crosses your limit.

Ecosystem section on landing — LangChain, CrewAI, n8n, LangGraph integration cards.

Integration guides in docs — LangChain KronvexMemory, CrewAI tool decorator, n8n HTTP node patterns.

Onboarding wizard — 4-step in-app guide shown on first login (API key, first memory, inject context). Dismissible.

DELETE /agents/{id} — endpoint was missing, now fully implemented with cascade memory deletion.

Stripe cancellation email — now sends a proper cancellation confirmation with end date + "reactivate" CTA.

Scale plan — memory capacity bar shows animated ∞ shimmer. Hourglass oscillates infinitely instead of filling.

Over-quota banner — shown automatically after downgrade when stored memories exceed new plan limit.

Stripe billing — Starter, Pro, Growth, Scale plans. Checkout, webhooks, plan upgrades fully automated.

Dashboard — complete redesign. Kronos hourglass banner, memory capacity bar, agents table, account page.

TTL decay — memories can now expire automatically. Background task runs every hour.

Session filtering — pass session_id to scope /recall and /inject-context to a conversation thread.

Confidence scoring — memories now have an access_count field incremented on every recall.

Onboarding email sequence — J+1, J+3, J+7 automated emails triggered on demo signup.

Blog — 4 technical articles published (memory types, RAG vs memory, production agent guide, quickstart).

CORS on 500 — global exception handler now always includes CORS headers to prevent silent frontend failures.

pgvector HNSW index — migrated from IVFFlat to HNSW. Recall latency p99 improved by ~35%.

POST /remember — store memories with content, type, session_id, TTL, and metadata.

POST /recall — semantic search over stored memories using pgvector embeddings.

POST /inject-context — one-call convenience: recall + format as LLM-ready context block.

GET/POST/DELETE /agents — full agent management with multi-agent support per API key.

Memory types — episodic, semantic, procedural. Scoped retrieval via memory_type filter.

Supabase auth — JWT-based auth via Google/email. Demo key created on first login.

EU hosting — all data stored in Supabase EU (Frankfurt). GDPR compliant by default.

Cloudflare Workers Sites — frontend deployed globally. Railway for FastAPI backend. Sub-40ms recall.

Publish to official registries — n8n community node registry · Dify plugin marketplace · MCP registry

Multi-tenant / organisations — team plans with shared agents and role-based access

Stripe live mode — production billing launch, real subscriptions