# Calmara

> Privacy-first AI productivity platform whose AI memory has a revision history you can prove. Every remembered fact is a subject/predicate/object triple with temporal validity, full version history, and provenance — inspectable, retractable, hard-deletable, and exportable in open formats (ICS, VTODO, JSON). Every assistant tool call is logged with the full input it sent. Tasks, calendar, projects and notes plug into Claude Desktop, ChatGPT, or any MCP client. BYOK, local LLM, or any OpenAI-compatible endpoint; EU-hosted; optional self-hosting.

## Product

- [Calmara — prove what your AI remembers](https://calmara.app/): AI memory with a revision history you can prove: every fact keeps its provenance, versions, and retractions. Plus tasks, calendar, projects and notes. BYOK, local LLM, or any OpenAI-compatible endpoint. Plugs into Claude Desktop and ChatGPT via MCP.
- [Pricing — Calmara](https://calmara.app/pricing): Calmara pricing: Free forever with 50 tasks, 20 notes, and 10 AI queries per month. Personal €8/mo for unlimited tasks, notes, AI, planner, auto-reschedule and Android widgets. Professional €15/mo for teams and power users.
- [Calmara for regulated professionals — GDPR-compliant AI productivity](https://calmara.app/for-regulated-professionals): Calmara for lawyers, therapists, healthcare consultants, NIS2-affected SMEs, and public-sector contractors. Self-host your knowledge and AI memory, run a local LLM with zero cloud egress, PII sanitization on every cloud call, GDPR-grade export.
- [Trust & security — Calmara](https://calmara.app/trust): Where Calmara stores your data (EU hosting on Railway, europe-west4), how AI calls flow (BYOK or local, PII redaction), encryption at rest, subprocessors, audit surfaces, and your GDPR controls.
- [Which AI assistants support custom MCP connectors? (2026)](https://calmara.app/mcp-compatibility): A checked, honest matrix: Claude, ChatGPT, Perplexity, Grok, Mistral Le Chat and local clients accept custom MCP connectors. The consumer Gemini app, consumer Copilot and Meta AI do not. Updated July 2026.
- [What's new in Calmara](https://calmara.app/changelog): Every Calmara release, newest first.

## Blog

- [Claude MCP server disconnected: causes and fixes](https://calmara.app/blog/claude-mcp-server-disconnected-fix): Why Claude Desktop, claude.ai, and Claude Code report MCP servers as disconnected, and how to fix each cause: PATH problems, config errors, the SSE content-type error, OAuth expiry, and the July 2026 spec change. With log locations and reproduced error strings.
- [A shared memory for every AI assistant: what the Obsidian-vault approach gets right, and where it stops](https://calmara.app/blog/shared-memory-for-ai-assistants): One person wired an Obsidian vault as shared memory for Claude Code, Codex, and a local LLM. The pattern is sound. Here's what a local vault can't reach, what it can't prove, and how the same idea works over MCP.
- [The EU AI Act's August 2 deadline came and went. Here's what actually happened](https://calmara.app/blog/eu-ai-act-august-2026-what-actually-happened): August 2, 2026 was the AI Act's big enforcement date, until a regulation passed six days earlier moved half of it. What switched on, what moved to 2027, who signed the transparency code, and what deployers actually need to do now.
- [Deleting a Claude chat no longer deletes what it learned from you](https://calmara.app/blog/deleting-claude-chat-does-not-delete-memory): Since July 2026, Claude's memory entries survive the conversations that created them: deleting a chat does not remove its memories. What Anthropic's docs actually say, how this inverts ChatGPT's problem, and what the monthly recap reads.
- [The July 28 MCP revision breaks both eras: what to fix in your server](https://calmara.app/blog/mcp-2026-07-28-spec-change): MCP's 2026-07-28 revision removes sessions, the initialize handshake, the GET stream and resumability. Legacy HTTP+SSE is deprecated, not removed. What that means for your server, with the exact status codes and headers.
- [Why your AI note app isn't end-to-end encrypted](https://calmara.app/blog/why-ai-note-apps-arent-end-to-end-encrypted): Every note app with real end-to-end encryption has no AI, and every one with AI can read your notes. Here's the technical reason, which vendors admit it, and the three ways out.
- [Google Keep vs Calmara (2026): the home-screen command center, and what Gemini reads](https://calmara.app/blog/google-keep-vs-calmara): Keep's widgets make an excellent Android command center. Wiring Gemini into it requires the one setting that allows training and human review. A verified comparison.
- [Suppressed or deleted? What actually happens when you remove an AI memory](https://calmara.app/blog/suppressed-or-deleted-ai-memory): Clicking 'forget' in ChatGPT, Claude, Gemini, or Copilot can mean three different things: hidden, deleted, or re-derivable later. What each vendor's docs actually say, what GDPR requires, and why retract and erase should be separate buttons.
- [Your AI has an audit log now (here's what it saw and did)](https://calmara.app/blog/your-ai-has-an-audit-log-now): Microsoft admits Copilot memory generates no Purview audit entries. ChatGPT, Claude, and Gemini don't log memory actions either. What an AI audit log should show, and what one looks like in practice.
- [How to migrate your ChatGPT memory (before it rewrites itself again)](https://calmara.app/blog/migrate-chatgpt-memory-before-it-rewrites-itself): A step-by-step guide to exporting what ChatGPT remembers about you and moving it somewhere you can actually audit. Covers the manual copy method, Claude and Gemini import, and what none of them preserve.
- [What did my AI believe about me in March? Time-travel queries for AI memory](https://calmara.app/blog/what-did-my-ai-believe-about-me-time-travel-memory): Most AI assistants can only tell you what they remember right now. A time-travel query answers a different question: what did the system believe about you on a specific past date, and when did that change? Here is what that looks like.
- [Motion vs Reclaim vs Calmara (2026): AI scheduling, with or without a privacy story](https://calmara.app/blog/motion-vs-reclaim-vs-calmara): Motion, Reclaim, and Calmara all auto-schedule your work in 2026. Verified pricing, which calendars and task tools each supports, where your data goes, and who can actually show you what their AI learned.
- [Notion AI vs Obsidian AI vs Calmara (2026): memory, privacy, and who owns your notes](https://calmara.app/blog/notion-ai-vs-obsidian-ai-vs-calmara): Three ways to add AI to your notes in 2026: Notion's hosted agents, Obsidian's build-it-yourself plugin stack, and Calmara's auditable memory. Verified pricing, privacy terms, and the honest tradeoffs.
- [ChatGPT's memory now rewrites itself. Here's what that costs you](https://calmara.app/blog/chatgpt-memory-rewrites-itself-audit-trail): OpenAI's June 2026 Dreaming update replaced ChatGPT's saved-memories list with a self-rewriting summary. What you lost, how Claude and Copilot compare, and what a real memory audit trail looks like.
- [How to install an MCP server in Claude Desktop (2026, step by step)](https://calmara.app/blog/how-to-install-mcp-server-claude-desktop): Three ways to add MCP servers to Claude Desktop as of July 2026: one-click .mcpb extensions, remote custom connectors by URL, and the manual JSON config. Steps, snippets, and troubleshooting.
- [The 6 controls your compliance team needs before approving an AI assistant](https://calmara.app/blog/ai-assistant-compliance-controls): A practical approval checklist for AI assistants: data flow, training exclusion, revocable keys, audit trails, deletion, and a no-cloud fallback.
- [The best local LLMs for private, regulated work (2026)](https://calmara.app/blog/best-local-llms-for-private-work): Five local LLMs worth running when client data cannot touch a cloud AI: verified hardware needs, licenses, and which ones can actually call tools.
- [How to run an AI assistant with zero cloud egress (Ollama + Calmara)](https://calmara.app/blog/run-ai-assistant-zero-cloud-egress): How to run an AI assistant where nothing leaves your machine: Ollama gives any app a local OpenAI-compatible API, and here is the full setup that uses it.
- [What is BYOK (Bring Your Own Key) AI, and why does it matter?](https://calmara.app/blog/what-is-byok-bring-your-own-key-ai): BYOK lets you plug your own OpenAI or Claude API key into an AI app, so your data runs under your account and terms. Here's what it fixes, and what it doesn't.
- [How to build an AI second brain you can actually audit](https://calmara.app/blog/ai-second-brain-you-can-audit): Most AI second brains are black boxes. Here's what an auditable one looks like: every remembered fact inspectable, sourced, and time-stamped, plus how to build one you can self-host.
- [EU AI Act enforcement starts August 2, 2026: what changes, and what doesn't](https://calmara.app/blog/eu-ai-act-enforcement-august-2026): On August 2, 2026 the EU AI Act's fines activate for AI providers, up to 3% of global turnover. Here's what actually changes if you only use AI tools, and what GDPR still doesn't cover.
- [How to audit what Claude remembers about you (and what it can't show you)](https://calmara.app/blog/how-to-audit-claude-memory): Claude has had persistent memory across all tiers since March 2026. Here's how to see what it remembers, what you can't audit, and how to export it.
- [Is MCP safe? The security risks AI assistants don't mention](https://calmara.app/blog/is-mcp-safe-security-risks): MCP servers can poison AI tool calls, leak credentials, and hijack accounts. Peer-reviewed 2026 research shows what's at risk and how to defend it.

## Optional

- [Author — Dan Hagen](https://calmara.app/blog/author): Founder profile and editorial background
- [Blog RSS feed](https://calmara.app/blog/rss.xml): Machine-readable feed of all articles
- [Privacy Policy](https://calmara.app/privacy): Data controller identity and processing details