Cadence by Stratium has three parts, and each one has a single job. Cadence reads and writes your calendar, and holds the plan and the record of your work. Claude provides the reasoning inside Cadence. Your own AI, if you use one, reads your inbox and writes what it finds into Cadence through MCP.
That is the architecture. The rest of this post is what it means for you, and for the person reviewing it in IT.
Key facts
- Cadence reads and writes your calendar in Microsoft 365 and Google, and reads Apple Calendar and ICS feeds.
- Cadence holds the plan and the record: work items, focus blocks, commitments, decisions, delegations and your priorities for the year.
- Cadence's built-in reasoning runs on Claude, through Anthropic's API.
- Cadence does not connect to your mailbox. Your own AI reads your mail through its own mail connector, and writes new asks into Cadence through the Cadence MCP connector.
- MCP is an open standard. You choose the AI, and you can revoke the connection at any time.
Why this matters
Most software sells the appearance of AI. The features carry the right words, the demos produce confident output, and the buyer is left to work out what the model actually does, what it does not, and where the data goes.
That is not a serious position for a tool you run your day on. You should not have to reverse-engineer marketing copy to find out whether your inbox is being read, where your commitments are stored, or which model sits behind the badge.
So we name the parts. The IT buyer can look each one up, the security team can read the safety posture, and you can see what is happening on your behalf without a translator.
The three parts
Cadence holds the plan and the record. Cadence reads your calendar and writes protected focus blocks back to Outlook or Google, where colleagues can see them. It holds your work items, the commitments you made and what other people owe you, your decisions and delegations, and your priorities for the year. When you close the day, it plans tomorrow around your meetings. A background job checks your focus time every three hours as meetings move. None of this needs a prompt. It runs in any browser, with native apps on Windows, Mac, iPhone, iPad and Apple Watch.
Claude is the reasoning inside Cadence. When Cadence fits your open work into the gaps between meetings, prepares a note for your next meeting, or lifts actions out of a transcript you have added, Claude does the reading and the writing. Cadence tells Claude what to do, in what order, and what a good result looks like. The execution is the model.
Your own AI reads your inbox. Cadence does not connect to your mailbox. If you want new asks in email to become work, you connect the AI you already use. Claude or ChatGPT reads your mail through its own Microsoft 365 or Gmail connector, under your account and your permissions. It then writes new asks and commitments into Cadence through the Cadence MCP connector. Set it as a morning task and it is done before you start.
Three parts, three jobs. Cadence keeps the structure and runs the calendar. Claude reasons inside it. Your AI brings in your mail.
What this looks like in a day
Theory is cheap. Here is the shape of an ordinary working day.
The evening before. You close the day in Cadence. Completed work is logged, unfinished items carry forward, and Cadence plans tomorrow around your meetings, with focus time for the work that needs it.
Before you start. If you have set a morning task in Claude, it runs. Claude reads your inbox and sent items since yesterday, writes new asks into Cadence as work items, and prepares your brief. Three Cadence skills for Claude, cadence-base, cadence-inbox-triage and cadence-daily-brief, give it the same steps every time.
When you sit down. Your morning brief arrives at the time you choose: the plan, the meetings that matter, anything at risk and the first piece of work. If you have an EA, they can adjust the plan before you read it.
During the day. Start a focus block and your Mac and iPhone go quiet while a timer runs the session. Complete or defer each item with one tap. When a meeting moves, Cadence re-plans around it.
None of this depends on you remembering to open an app.
Why we built on Claude
Cadence needed a reasoning engine for planning, meeting preparation and transcripts. Two things decided it for Claude.
Skills. Cadence is not a chat product. Its work is routine: the same steps, in the same order, every day. Claude's Skills let those steps be written down once and followed the same way each time. The Cadence skills for Claude, which run your inbox triage and your morning brief, are built that way.
Safety posture. Your working data is sensitive. Calendar entries reveal strategy. Commitments reveal judgement. We wanted to build on a model whose maker publishes its safety thinking and holds its API customers to it. Anthropic does. We picked Claude in part because it is the company we want our customers' security teams to read about.
Building on Claude does not mean you have to use Claude yourself. Cadence's built-in AI works on its own, and if you already use ChatGPT, you can connect that instead.
What MCP is, and why it matters
The Model Context Protocol is an open standard for connecting AI tools to the systems they work with. Anthropic proposed it in late 2024. By late 2025 it had passed ten thousand active public servers and ninety-seven million monthly SDK downloads. Claude and ChatGPT support it, and so do Gemini Enterprise, Cursor and VS Code. It is the closest thing the industry has to a USB-C standard for AI connectors.
Cadence runs an MCP server. That does three things for you.
You choose the AI. Claude and ChatGPT both connect to Cadence through the same server. You keep the AI you already use and pay for, and your plan becomes part of what it can see.
It works in both directions. Your AI can read your plan, your calendar and your work items, so it knows your week before it answers. It can also write back: new work items, follow-ups, delegations and focus time. Read requests change nothing. Writes run under your own Cadence identity, so they appear in the activity feed as if you had made them yourself.
It is yours to revoke. The Cadence connector uses OAuth 2.1 with PKCE, rotating refresh tokens, and tokens bound to the Cadence server. There is no long-lived API key to copy or leak. Disconnect it in Cadence and access stops.
Where your data lives
Your mail stays where it is. Cadence never connects to your mailbox, so your inbox is not copied into Cadence. What reaches Cadence from your email is the work item your AI writes, not the mailbox behind it. Your mail stays between you, your mail provider and the AI you chose to connect.
Cadence stores what it needs to run your day: your plan, work items, focus blocks, commitments, decisions, priorities, the calendar events it syncs and any transcripts you add. All of it sits on Microsoft Azure West Europe by default, with each organisation's data kept separate and encrypted in transit and at rest. Sovereign hosting is available for organisations that need it. Your content is not used to train models.
The IT due diligence answer
The short version, for an IT leader asked to approve Cadence:
- Cadence reads and writes the calendar through Microsoft 365 and Google, and reads Apple Calendar and ICS feeds.
- Cadence does not connect to mailboxes. Users who want email turned into work connect their own AI, which reads mail through its own connector under the user's permissions.
- Cadence's built-in reasoning runs on Claude through Anthropic's commercial API. Content is not used to train models.
- AI tools connect to Cadence through MCP, using OAuth 2.1 with PKCE, rotating refresh tokens and tenant isolation.
- Cadence data is hosted on Microsoft Azure West Europe by default, tenant-isolated and encrypted. Sovereign hosting is available.
- SOC 2 Type II is in progress.
The full detail is on the Trust and security page.
Why we do not pretend to be an AI company
Stratium is not an AI company. Cadence is the structure of a working day, expressed in software: the plan, the calendar, and the record of what you promised and what you did. The reasoning is done by an engine we did not build, and we are not embarrassed to say so.
That position holds up in three ways.
Commercially, buyers are tired of products that sprinkle "AI" across the marketing and produce mediocre output. Naming the model is a credibility signal, not a risk.
Technically, as Claude gets better, Cadence gets better. As MCP spreads, more of the AI tools you use can work with your plan. We put our own engineering into the structure, because that is the part that is ours to solve.
Practically, most people now have an AI of their own. The useful question is not which AI a product has built in. It is whether the product works with the one you already chose.
Cadence holds the plan and runs the calendar. Claude does the reasoning inside it. Your own AI brings in your mail. That is the architecture, and that is what we will keep saying.
Frequently asked questions
Is Cadence built on Claude? Yes. Cadence's built-in reasoning runs on Anthropic's Claude, through Anthropic's commercial API.
Does Cadence read my email? No. Cadence connects to your calendar, not your mailbox. Your own AI can read your inbox through its own mail connector, and write new asks into Cadence through the Cadence MCP connector.
Can I use Cadence with ChatGPT instead of Claude? Yes. ChatGPT connects to Cadence through the same MCP server, on Business and Enterprise plans once an admin turns on custom connectors. Cadence's built-in AI runs on Claude either way.
What is MCP? The Model Context Protocol. An open standard proposed by Anthropic in 2024, and now supported by Claude, ChatGPT, Gemini Enterprise, Cursor and VS Code. It is how AI tools connect securely to the systems you already use.
Does Cadence store my data? Cadence stores what it needs to run your day: your plan, work items, focus blocks, commitments, decisions, priorities, the calendar events it syncs and any transcripts you add. It does not store your mailbox, because it never connects to it.
Where is Cadence hosted? Microsoft Azure West Europe by default, with each organisation's data kept separate and encrypted in transit and at rest. Sovereign hosting is available. SOC 2 Type II is in progress.
Who can use Cadence? Anyone whose calendar fills faster than they can manage it, on their own or with an EA. Cadence runs in any modern browser, with native apps on Windows, Mac, iPhone, iPad and Apple Watch.
For how Cadence runs a day in practice, see How it works. For the full security and data architecture, see Trust and security. To set up your morning brief in Claude, see Connect Cadence to Claude.