Personal software shaped my career—building tools where teams live every day taught me that distribution and trust matter as much as features. Today applies that lesson to AI: the agent has to be something people permit into calendar and mail, not a toy they open once and forget.
Why We're Building an AI Personal Agent
The most capable AI in history is already in everyone's pocket. It can plan, research, remember, write, and act. And yet, for almost everyone, it just sits there waiting to be told what to do. Power arrived before relationship. People have a genius on call and use it like a search box—not because they are wrong, but because the tools were built for answers, not for presence.
We think that is backwards. A capable AI personal agent should pay attention, remember what matters, notice change, and show up before you ask. Not a smarter chatbot—a companion that closes the gap between what AI can do and what your life actually gets from it. IBM's framing of assistants versus agents puts it plainly: assistants react; agents pursue goals with tools and plans that can continue without you re-prompting every turn.
Three commitments anchor Today: it knows you, it acts before you ask, and it walks with you toward the life you want.
We are not trying to win a benchmark on a single prompt. We are trying to ship the kind of software you leave running—because it compounds context the way a great chief of staff compounds, quietly, over quarters. That ambition is why we call Today an AI personal agent in public, and why we wrote a separate definition post instead of hiding behind a vague "assistant" label. For the vocabulary—memory, initiative, confirm-before-execute—read what an AI personal agent is.
It Knows You — Memory You Can Trust
An agent that does not know you can only answer you; one that knows you can help. Most AI meets you as a stranger every time—the tab closes and the context vanishes. That is not an assistant or an agent. That is a vending machine.
Being known only works if you stay in control. Today holds memory you can read, edit, and erase—transparency as a design requirement, not a footnote. We did not want "personal context" to mean a black box the vendor owns and you hope is right. We wanted rows you can audit the way you audit a contact list: see the fact, fix the fact, delete the fact, move on.
In early Beta, people tell us the product "gets them." We treat that as the bar: not clever replies, but continuity that compounds. Trust is also revocable. Every connector you add should be removable without losing the core agent—conversation, memory, briefs. That design choice reflects how personal agents must behave: present by default, invasive never.
We are not building the loudest assistant in the feed. We are building the one you keep—because it remembers, prepares, and waits for your yes. Ecosystem assistants are racing toward the same problem—Apple's Siri AI promises personal context across messages, mail, and photos—but a dedicated personal agent still has to earn trust through inspectable memory, not only retrieval magic behind the glass.
If that resonates, try Today on Mac, iOS, or Android — Today is live in public beta now.
It Acts Before You Ask — Agency Without Spam
Knowing you is half the job. What you feel daily is what the agent does with that knowledge—quietly, before you ask. A good friend does not wait for a formal request before saying "don't forget tomorrow's thing." That is the proactivity we mean: timed to your day, not broadcast because the model can.
Chat is a terrible place to keep a life—everything scrolls away. A personal agent should surface what matters in views you return to, not bury it in threads you dig for. And it should work while you live: long-running tasks that stay alive across days instead of dying when a chat session ends.
We hold a strict notification bar—details in the product post—because an agent that pings everything gets muted, and then it helps no one. The comparison that matters here is not "more alerts versus fewer alerts" in the abstract; it is signal tied to your day versus noise tied to model capability. Classic assistants stay reactive by design; personal agents add initiative—and must pay for it with restraint.
Proactivity is not autonomy. Today will not send email, move calendar events, or message others without your explicit confirm. That line protects trust and keeps the product on the right side of "personal agent" versus "autonomous bot." If you want full autopilot, Today is the wrong tool today—and probably the wrong mental model for software this close to your identity. For how assistant-style reactive help compares to agent-style continuity, see our guide on AI personal assistant vs AI personal agent.
It Walks with You — Life Arcs, Not Errands Only
The point was never to clear today's tasks slightly faster. It is to move your life forward across months—fitness, learning, savings, relationships, craft—not one errand at a time. Productivity software optimized the working slice of you. We disagree.
Early Beta users use Today for work and for health routines, language practice, and long projects that do not fit a single prompt. That is the agent we want: one thread held across seasons. Lifestyle health workflows—meal planning, sleep rhythm, training—live on our Healthcare hub with clear lifestyle support, not medical diagnosis boundaries.
Getting in shape. Learning photography. Preparing for a baby. These are arcs, not checkboxes. An AI personal agent should remember where you are going and nudge you toward it—not reset every Monday.
Consider a founder preparing for a launch while traveling: the agent should connect flight days, sleep debt, and promised deliverables without a daily re-brief. Consider a parent returning to fitness after injury: memory should hold clearance timing and prior load, suggesting a lighter week before you overreach. That is a different job from delegating a deck or spreadsheet to a work sandbox—and the category split matters when buyers conflate them. Read AI personal agent vs work agent when you need whole-day continuity separated from Cowork-style deliverable factories.
Capability Without the Hype
We do not lead with the machine, but warmth is not a substitute for follow-through. Under the surface, Today is a real agent stack: model orchestration, planning, memory retrieval, task queues, cross-device sync. We invest in internal evaluations so improvement is measurable—not merely claimed.
The unglamorous work matters: infrastructure that stays responsive, jobs that finish instead of stall, sync that lands on the right screen. Knows you and acts before you ask are demos until reliability makes them habits. Research on long-horizon AI systems keeps converging on the same insight: models are rarely the bottleneck anymore; state, tooling, and evaluation are. Anthropic's writing on building effective agents emphasizes disciplined scoping and human checkpoints—the same instincts we apply to calendar and mail.
We also believe evaluation beats adjectives. Internally, we test planning, memory retrieval, and connector success rates on scenarios drawn from real Beta days—not synthetic trivia questions. That discipline is how an AI personal agent earns the word agent instead of borrowing it for a landing page.
We'd Rather Be Honest Than Impressive
Today is not finished. An agent this close to your day earns trust in small things—the right time, the right fact, follow-through that does not drop—so we are shipping to people who want to grow with us and improving in the open. Features may change during Beta; the product is free for now (Terms). The goal was never more time in an app; it is more of your life back. When Today works, you think about the trip, the skill, the people—not the software.
Beta users shape the roadmap directly—feedback goes into memory, brief, and connector priorities on each release. That loop is intentional: personal agents should co-evolve with the people they serve, not ship as frozen demos.
Today is your day.
Conclusion
If the idea of an AI personal agent—memory you control, help that shows up early, arcs that stretch beyond one afternoon—matches how you want to live with AI, start at the landing page or get started today.
We are early. The agents that win the next decade will not be the ones with the flashiest demo—they will be the ones people trust with Tuesday. That is the standard we are building toward.