Common questions, direct answers.

No. Your data is stored on your own dedicated Mandaire instance. The AI you use, Claude, ChatGPT, or Gemini, receives only the specific context you choose to pass via MCP tools. Nothing is sent in bulk to any provider.

When you ask a question through your AI, Mandaire runs the query locally, applies your disclosure rules, and returns only the allowed result to the AI. The AI sees that result. It does not see the rest of your data.

No. Your data is never used to train any model. It lives on your dedicated Mandaire instance and is not shared with any other user.

Yes. Export runs today. Request a full export of your graph, inferences, and raw corpus from the privacy request page. The architecture does not depend on us holding your data: the graph, the inferences, the calibrated record, all of it is yours. Migrate to self-hosting or another provider at any time.

Only who you authorize. A deterministic disclosure gate runs before any query reaches your AI: scoped by person, topic, and context, with every decision logged to the audit record.

The same question from two different people can return structurally different context. Not because Mandaire withholds arbitrarily, but because what is appropriate to share with your investor is not the same as what is appropriate to share with your spouse. The disclosure rules are yours. They run before the AI sees the result.

LLM memory stores what you tell one AI in one session. Mandaire builds a cross-source knowledge graph: 20 years of email, every message thread, your calendar, your files, resolved and calibrated across all of it simultaneously.

Three structural gaps that LLM memory cannot close, regardless of scale:

Deduplication. "Alex" in your email is a different person from "Alex" in your messages. LLM memory does not resolve this. Mandaire does, using co-occurrence patterns, shared identifiers, and communication history.

Recency weighting. A one-time mention from five years ago should not carry the same weight as a recurring thread from last week. LLM memory does not distinguish them. Mandaire calibrates against the full corpus.

Per-recipient disclosure. ChatGPT, Claude, and Gemini cannot return a different answer to the same question for two different recipients. The disclosure primitive does not exist in the generative paradigm. Mandaire applies this before the AI sees the result.

No. A system prompt is a static paste. It does not update between sessions, does not resolve entities across sources, and does not apply per-recipient disclosure rules.

Mandaire is queried at the moment your AI needs it. The result is current, scoped to the question, and pre-filtered by your disclosure policy. The AI receives a targeted answer, not a dump of everything.

Platform incentives prevent it. Apple cannot ingest Gmail or Meta's social graph. Google cannot reach iMessage or WhatsApp. Each platform owner's incentive is to deepen their own silo, not bridge a competitor's.

The cross-vendor fragmentation is not a bug in the landscape. It is a structural property that competitive incentives actively maintain. Only a user-authorized neutral third party can span all silos simultaneously. The harder the platforms compete, the more the joining problem belongs to a neutral layer.

Claude, ChatGPT, and Gemini. Mandaire connects via the MCP protocol: add one URL to your AI client and it reads from the graph. No separate app, no interface to learn.

Any AI that supports MCP can connect. The graph is the same regardless of which AI you use.

It tells you. Confidence is part of every output: the system surfaces what it knows, how many sources it has seen, and where the gaps are. An answer based on thin data arrives flagged as thin.

A confident-sounding answer built on two data points is a different kind of problem. Mandaire is calibrated to avoid it.

For mandaire.app (the personal context layer): no. You connect your accounts and AI through a guided setup. No code required.

For mandaire.dev (the builder tier): yes. You are connecting an AI you build to a personal knowledge graph via MCP. That is a technical integration.

Mandaire is in private beta. Request early access and we will follow up with next steps. We are prioritizing people with a clear use case and a clear sense of what they want to connect.