Prove what your AI remembers

Every fact Calmara's AI learns keeps its revision history: where it came from, when it changed, what replaced it. Retract it, delete it permanently, or export the lot. Tasks, calendar, projects and notes underneath, plugged into Claude Desktop or ChatGPT.

Free forever. No credit card required.

AI memory you can prove

Every fact Calmara's AI remembers is stored as a plain-English subject/predicate/object statement with a temporal validity range — "works at Acme, true since January 2025". You can read every entry, edit it, retract it, or look back at what the AI believed last quarter. A pending-review queue means nothing is remembered without your explicit approval, and every fact keeps its history: when it became true, when it changed, and where it came from. That's the audit trail cloud AI assistants don't give you.

Works inside Claude Desktop and ChatGPT

Calmara exposes a Model Context Protocol (MCP) server, so Claude Desktop, a ChatGPT Custom GPT, Cursor, Continue, or any MCP-compatible client can read your memory, tasks, calendar, projects, and notes. Ask "what did I decide about the Nordics rollout in April?" and the answer comes from your actual records — not a hallucination. Setup is one copy-paste, secured with OAuth 2.1, and available on every tier. The same memory follows you across Calmara, Claude Desktop, and ChatGPT.

This is what "auditable" looks like

Not a promise in a privacy policy — three screens in the product, from a real account.

Every version of every fact

Open the history on any fact: what it used to say, when it changed, which assistant recorded it, the sentence you said that caused it, and what replaced it.

Calmara's fact history panel showing three versions of a role fact — Founder & Operator (current), Head of Product (retracted), Engineering Manager (retracted) — each with validity dates, how it was recorded, and what replaced it.

Every call an assistant made

The AI activity log lists each tool call against your account — which tool, from which assistant, how it authenticated, whether it worked — and expands to the exact input it sent.

Calmara's AI activity log listing tool calls with MCP and Pulse badges, timings, a failed call, and one entry expanded to show the raw JSON the assistant sent.

Nothing is remembered without you

Facts an assistant proposes, or that you import from ChatGPT, wait in a review queue. You approve or reject each one. Rejections stay in the history as retracted.

Calmara's pending review queue with three facts imported from ChatGPT, each with Approve and Reject buttons.

Every AI action leaves a record

Settings shows an activity log of every tool call a connected assistant made against your account, with the full input it sent. The nightly scheduler keeps its own log: which task it moved, from which date to which. Any fact can be retracted, which keeps it in the history as superseded, or hard-deleted, which erases the record entirely. If you cannot check it, it is not a control.

Bring your own LLM — even a local one

Use Gemini, Claude, OpenAI, or Z.AI with your own API key, point Calmara at any OpenAI-compatible endpoint (Azure OpenAI in an EU region, Scaleway, OVHcloud, or a vLLM server you run), or run a local model on your own machine through Ollama. A local LLM means AI features work with zero data egress to any cloud provider, and bringing your own key lifts the AI query cap on every tier, including Free. Your keys, your model, your call.

A complete productivity workspace

Underneath the memory layer, Calmara is a full tasks-calendar-projects-notes app. Google Calendar, Microsoft, and Todoist sync into one agenda, with work/private context separation that keeps your evenings yours. Projects get Kanban boards, Gantt charts, dependencies, and custom fields. The knowledge workspace links notes like Obsidian, keeps daily journals, and works offline. Hybrid search finds things by meaning, not just keywords — ask for "Q3 planning" and get the strategy doc even when it never says "Q3". There's an Android app with home-screen widgets, too.

Your data, your rules

You choose where your data lives: keep it hosted by us, or run your knowledge and AI memory on your own PostgreSQL with a one-command Docker bundle. Structured identifiers — emails, phone numbers, card and ID numbers, IP addresses — are redacted before prompts reach a cloud LLM. Analytics are cookieless, there are no ads and no third-party trackers, and everything exports in open formats on every tier: calendar.ics, tasks.ics (VTODO), and a schema-versioned JSON file with the lot.

Who Calmara is for

Calmara is built for people whose compliance officer — or conscience — won't let them paste client data into a cloud chatbot: lawyers, therapists, healthcare and security consultants, and public-sector contractors (see Calmara for regulated professionals). It works just as well for solo founders and knowledge workers who want an AI that actually remembers their work — and lets them prove what it knows. For the long-form version of how we think about AI memory, privacy, and MCP security, read the Calmara blog.