Mandaire is the private context layer underneath the AI you already use. It joins your email, messages, calendar, notes, files, photos, and AI conversations into one source-backed picture, and decides what each assistant gets to see. Keep using the walled gardens. Stop being stuck inside any one of them.
The shape of the problem
Gmail knows one part of your life. Outlook knows another. iMessage, WhatsApp, calendar, contacts, photos, notes, files, browser, ChatGPT, Claude, Gemini — each holds a fragment. None of them connect.
Provider memory does not fix this. ChatGPT remembers what happened inside ChatGPT. Gemini sees what Google can see. Claude sees what you paste. Each one will get better at remembering its own slice. None of them will become the layer across all the others. That is not a bug they are going to fix. It is how their business works.
Mandaire sits above the walls. It builds your private context graph on infrastructure you own, then exposes only the right context to the AI or app you are using. The walled gardens stay walled. You stop being stuck inside any one of them.
In practice
Three things Mandaire does that the AI you already use cannot.
Before a high-stakes meeting
You have a 10:00 with Marcus at Vertex. Mandaire knows the last email thread, the coffee with him a year ago in your photos, his kids' ages from his birthday-party photos, his wife's August text mentioning the school year, and a concern about the CMO that a former colleague raised off the record last month. It produces a five-minute brief with sources, caveats, and a clear note on what should stay out of writing.
Five minutes of context that would take you an hour to assemble. The full brief is further down.
When commitments slip across channels
You promise Lin on WhatsApp Tuesday you will send the draft by Friday. Wednesday she emails about a different angle. Thursday she texts a quick follow-up. Mandaire connects all three, reminds you the draft is still open, and offers a single reply that covers everything. Without it, one of the three threads gets dropped.
Nothing falls through because nothing is in only one place.
When the message needs to land right
A former colleague emails asking if you would be a reference for a job he is after. You want to help, but you have a real reservation about the role you would never put in writing. Mandaire drafts the reply, drawing on what you have actually said about him across email, calls, and your notes. Warm to the friendship; specific about the strengths you can endorse; raises a question worth his thinking about himself. The concern a mutual friend shared with you last month stays out, because you told Mandaire that conversation was off the record and it remembered. You change one word and hit send.
The careful version, in the time you have.
When a friend has gone quiet
Half an hour before lunch with Maya, Mandaire surfaces the right Maya, not the three other Mayas in your contacts. Last time you saw her was her summer dinner two months ago. In messages and email since: the move to Marin, the boys starting middle school, her mother's health. The obstetrician contact you said you would send has been open since spring. Your reply cadence with her has slowed compared with the prior six months — observed, not interpreted. Worth noticing before lunch.
Observational, not psychologizing.
Why this is different
Mandaire separates reasoning from rendering. A private reasoner runs inside your own Mandaire and is the only thing that sees your raw data. The AI you chat with — Claude, ChatGPT, Gemini, a local model — only receives a disclosure-filtered briefing, the way a chief of staff would brief you before a meeting. You get the useful context without handing your full life to every AI provider.
Mandaire joins your email, messages, calendar, contacts, photos, notes, docs, files, and AI conversations into one source-backed graph. The same Sam in your inbox is the same Sam in iMessage and the Sam at your daughter's wedding — not the other Sam from your old job. Every claim points back to where it came from.
The context follows you, not your provider. Ask Claude about a relationship and it answers from your actual history. Switch to ChatGPT next week, same answer, no re-priming. Run a local model for sensitive work, still the same picture. Your memory stops being trapped inside one vendor's product.
Not all context should travel. Your therapist notes, family messages, work email, and private AI conversations should not all go to the same place. Mandaire decides per person, topic, context, and surface what each request is allowed to see — and logs every disclosure so you can audit it later.
What the output actually looks like
Below is what Mandaire actually produced for the Marcus example above. Notice the parenthetical source note ("found in your photos; not in your calendar"), the explicit sources at the bottom, and the caveats marking what is inferred versus directly stated. Mandaire shows its work. No memory feature inside a single provider can join sources that disagree, because none of them reach across the walls.
Mandaire surfaces what you have been carrying. For some of us that is relief; for some it is uncomfortable. You can ask it to back off on a topic, a person, or a stretch of time. Use it when the message matters; not every moment needs preparation. The product is built to be useful, not punishing.
vs single-provider AI memory
Anthropic, OpenAI, Google, and Apple are all building memory into their assistants. Those features are useful. They will get better. They will also stay inside their own walls. Mandaire is a different layer of the stack.
Remembers what happened inside one provider. Lives on the provider's servers under the provider's privacy policy. Personalizes one assistant. Stores facts, not source-backed claims. One memory policy for the whole memory. Hard to audit; hard to correct; not portable when you switch providers.
Sees across email, messages, calendar, contacts, photos, notes, files, and AI conversations together. Runs on infrastructure you own, encrypted with a key only you hold. Gives the same source-backed picture to every assistant you use. Per-audience disclosure rules. Sources, caveats, corrections, and full export on every claim. Switch providers any time; the memory comes with you.
The natural next step is for provider memory and Mandaire to work together. Plug Mandaire into ChatGPT, Claude, or Gemini, and the provider's memory keeps doing what it does well inside its own surface. Mandaire holds the picture that no single provider can.
Where to start
The fastest first step is dropping in a takeout export of your AI conversation history. Years of self-articulation in your own words land in minutes from a ChatGPT, Claude, or Gemini export. You get something useful the same day, before you have connected anything else.
From there, add accounts at your own pace. Email and calendar deepen the picture quickly. Messages add cadence and tone. Photos pin people and events to time and place. Each new source compounds the others; even one is useful on its own. Most users have a working assistant within an afternoon and a serious picture within a week.
No native Mandaire app to install. Most providers (Google, Microsoft, Meta) connect over standard OAuth. Apple sources (iMessage, Photos, Notes) need a small local agent on a Mac you own, since Apple does not offer consumer OAuth for these.
Where you use it
You do not have to add a new app to your day.
Through the AIs you already use. Claude, ChatGPT, Gemini, or a local model. Mandaire is connected as a private context source. Switch providers any time and your Mandaire comes with you.
Private chat with the reasoner directly. For the things you do not want filtered: corrections, raw thinking, the quick "remind me what I said about X." Encrypted, just you and your data. No external AI in the loop.
Drafts and notes inside the apps you already use. Mandaire drafts replies in Gmail, suggests responses in iMessage, writes into Apple Notes or Obsidian, drafts documents in Google Docs, adds calendar events, captures commitments. Everything outbound is read-only or draft-and-approve by default; nothing is sent on your behalf without your sign-off.
A day with Mandaire
7:15 AM. A short morning brief lands in your private chat: who is on your calendar, what shifted overnight in email and messages, who you owe a reply to today. You read it on your phone over coffee. Two minutes.
10:33 AM. Thirty minutes before your eleven o'clock with a partner you have not seen in a while, the pre-meeting brief lands the same way. The right person, the last conversation, what they care about, what is open. You skim it on the way to the call.
2:48 PM. You open Claude on your laptop to ask about something unrelated. You ask "what was the takeaway from the GTM thread with Ana last week" and Claude answers in one reply: who Ana is across your email and Slack, the thread ran from Tuesday to Friday, what you concluded. You did not paste anything in. Mandaire was plugged in.
6:30 PM. An end-of-day brief: what you committed to today, what you set up for tomorrow, what to think about overnight. Same private chat. Two minutes.
In between, when you ask any AI you use about anyone or anything in your life, the answer is grounded in what you actually said, not what the AI guessed. The AIs you already use stop guessing. You do not have a new app to check.
Who this is for
If you use more than one AI assistant, if important context is spread across email, messages, calendar, notes, files, and AI chats, if you regularly prepare for meetings by digging through old threads, or if you have relationships and commitments that span channels — Mandaire was built for you.
Concretely: founders, executives, chiefs of staff, investors, operators, consultants, lawyers, researchers, and AI power users in their first 90 days of any new role tend to feel the pain first.
Who this is not for, yet
If all you want is memory inside one assistant, use that assistant's built-in memory feature. If you do not want to connect personal data, Mandaire is not useful yet. If you want a fully hosted SaaS where the vendor owns the infrastructure and the data, Mandaire is intentionally not that — sovereignty is the architecture, not a marketing claim.
Private
A promise not to look at your data is not the same as being unable to look at it. We believe you own your data, not big tech. Mandaire runs on a server in your name, in your cloud account, encrypted with a key only you hold. We do not have the bill. We do not have the key. We do not sell your data, train models on it, or share it.
In managed hosting, our operations team has limited, logged access for setup, patches, and break-fix. That is fully avoidable by self-hosting on your own hardware. The encryption module that handles your data is on track for open source with reproducible builds and a third-party audit in Q3 2026. Until then, we publish the schema, the threat model, and the architecture in public so you can evaluate them now. See architecture →
Cost
Mandaire is free for individual use during private beta. The cloud server it runs on is paid by you directly to your cloud provider; a small server typically runs $5 to $20 a month. The AI you choose — Claude Pro, ChatGPT Plus, Gemini Advanced, or a local model — is your subscription and your bill. Total typical, all in: under $50 a month.
A small flat management fee for the operated stack (setup, updates, security patches, monitoring, backups) will apply when Mandaire moves out of beta. The core self-hosted software will remain free for individuals. Paid professional and team plans will be announced before general availability.
Common questions
No. Mandaire sets up your server for you, the same way managed hosting sets up a website. You connect your accounts the way you would connect them to any app. There is no Linux to manage and no command line to learn.
Most users start with their AI conversation history, which lands in minutes from a takeout file and produces obvious value the same day. Connecting your first source takes a few minutes. Connecting all of them, longer, but you can do it at your own pace.
Yes. You reach Mandaire on your phone through the AI you already use, through your private-chat app, or through your connected messaging and email apps. There is no separate Mandaire app to install. Most providers (Google, Microsoft, Meta) connect via standard OAuth, no extra agent needed. Apple sources (iMessage, Photos, Notes) need a small agent on a Mac you own, since Apple does not offer consumer OAuth for these.
Yes. Everything Mandaire holds is exportable on demand. You can also ask it directly. "What do you have on me about Sam?" returns the actual sources, not a summary you have to take on faith. Every claim has a citation. Every inference is marked as inferred.
Organizational accounts (your employer, university, non-profit, agency) often do not permit personal tools to connect. If yours does, you can plug it in. If not, most people connect their personal accounts only and leave the organizational world separate. The two need not meet.
It does not matter. Mandaire works with what you have, on whichever sources matter most to you. A recent graduate with photos, messages, and a few hundred AI conversations gets a useful picture immediately. A long-time professional with two decades of email gets a deeper one. The picture deepens as you use it; the product does not gate on a starting volume.
You tell it once. "That is the wrong Maya." "I have not gone quiet on Jen, we just moved to Signal." Mandaire takes the correction, applies it across every chat and brief, and remembers. Wrong calls are normal early on, especially with relationships that do not have much recent data. The picture sharpens with every correction. Every brief and every claim has a source you can check.
Those features are useful inside their own surface. They do not, and structurally will not, see across providers. Mandaire is the layer underneath: one source-backed picture across all of your sources, available to whichever assistant you happen to be using.