Dwight · Your Agentic Eisenhower Matrix
Make important work easier to do.
Dwight reads your calendar, inbox, documents, and work signals, separates importance from pressure, and helps you spend more of the day on what actually matters. It does that by clearing drag, protecting time, staging actions, and learning what gets you into the work. Every write action still waits for your approval.
Dwight reads live work signals, makes the important work easier to enter, and never sends or changes anything without your sign-off.
The system behind it
The matrix is the judgment engine, not a decorative framework.
Dwight uses the Eisenhower Matrix because the real question is not just what to do next. It is how to spend more time on work that matters. Urgency and importance are scored from live signals, then Dwight adds the layer most matrix tools miss: it helps clear the friction around the important work so it is easier to start and protect.
(Yes, the product is named after him.)
Do Now
Urgent + important. Handle it today.
Protect
Important, not urgent. Defend this time.
Respond
Urgent, less strategic. Delegate or batch.
Ignore
Neither. Let go.
How Dwight works
Step 1
Sees the real work
Connects to Calendar, Gmail, Drive, Slack, and Toggl. Pulls the commitments, waiting replies, active documents, logged time, and pressure signals already shaping the day.
Signals pulled now
→ Building matrix from 23 signals...
Step 2
Separates signal from noise
Claude judges every item on time pressure and strategic importance. Where it lands on the matrix is the judgment: what is loud, what is consequential, and what deserves defended attention.
Scoring in progress
Reply to client proposal
→ Placed in DO NOW
Step 3
Clears drag around the work
Routine work gets drafted, blocked, logged, or routed so the important work is easier to enter. Everything staged. Nothing moves until you approve.
Approval queue
Draft reply to Sarah re: proposal
gmail draft
Block Tue 14:00 for course work
calendar
The agentic layer
What Dwight actually removes for you. Every item below lands in an approval queue first.
Let Dwight Draft
Do Now cardEmail reply written from the thread. Staged in Gmail. You approve, edit, or skip before anything sends.
Block the time
Protect cardCalendar event created at the next open gap. First move drafted and ready when you open the workspace.
Stage for delegation
Respond cardReply drafted and routed to the right person. Or batched into a single review queue.
Log the work
Any cardToggl entry started and assigned to the correct project. No manual project-switching.
Intelligence and memory
How Dwight gets sharper over time and helps shift more of your day toward work that matters.
Multi-source scoring
Eight data sources are fused in parallel before every score. Claude judges the result. A gravity layer then applies a bounded post-pass from additional signals such as drag history, goal anchors, and personal priors. The point is not just ranking tasks. It is identifying where Dwight can remove friction and where only your presence will move the real work.
Fingerprint memory
Every task shape is tracked across sessions. Dwight learns which cards you finish quickly, which ones become avoidance loops, which kinds of staging actually help, and which sources create noise. Memory is only useful here when it changes the next judgment or removes restart friction.
The journey to deep work
Each day Dwight tracks your state: reactive, mixed, or clear. The goal of the whole system is to move you toward deeper, higher-impact work by making that work easier to start, easier to protect, and harder to quietly abandon. After a Protect block, Dwight checks whether you used it and adjusts the next day's scores accordingly.
Trust and privacy
What is actually implemented. No marketing claims here.
What is verified and live
- ✓
Row-Level Security on every Supabase table
All sensitive tables are locked to service_role only. The anon and authenticated roles are explicitly revoked at the database level.
- ✓
No passwords stored
Authentication is Google OAuth 2.0 via NextAuth v5. Only OAuth tokens are held, managed by the auth library.
- ✓
Rate limiting fails closed
Upstash Redis sliding-window rate limits on all API routes. In production, a missing Redis binding denies requests rather than allowing them through.
- ✓
Approval queue on all write actions
Email sends, calendar creates, Toggl entries, Slack messages: all staged and reviewed before execution. Enforced at the database level with a state-machine constraint.
- ✓
Your data, your database
Signals are stored in a Supabase instance you control. The builder has no access to your data.
- ✓
LLM inputs not used for training
Dwight uses the Anthropic API. Per Anthropic's enterprise terms, API inputs are not used to train models.
What is not yet there
Dwight is a personal tool built by one person. These are the honest gaps. They matter more as usage grows beyond a single user.
- △
No formal privacy policy page
There is no /privacy or /terms page. One can be written and published.
- △
No GDPR right-to-erasure API
Data deletion requires direct Supabase dashboard access. A DELETE /api/user route needs to be built.
- △
Broad OAuth scopes
Calendar, Gmail, and Drive permissions are granted at the full read/write scope. Granular scope selection is not surfaced to users during sign-in.
- △
No SOC 2 or ISO 27001
No third-party security certification exists. This relies on Supabase's own infrastructure compliance.
Frequently asked
What exactly does Dwight read?
Google Calendar, Gmail, Google Drive, Slack saved items, Toggl time entries, and Granola meeting notes. You connect what you want. Dwight reads and builds the matrix. It never posts, sends, or edits anything without your explicit approval.
How is this different from a to-do app?
Dwight builds the list for you, from your actual signals, not from tasks you manually enter. The matrix position is the judgment: urgency multiplied by importance, scored live from your sources. You drag to disagree; Dwight learns from the drag.
How does Dwight learn over time?
Every drag teaches it. Cards you move away from strategic goals get flagged as avoidance. Items you repeatedly skip get escalated. After a few weeks Dwight has a precise model of your actual patterns and uses that to pre-score new cards.
Who is this built for?
Operators, consultants, and executives who juggle multiple clients and projects. People whose work lives in email, calendar, and documents, not in project management software. People who want an AI that acts, not just advises.
Calendar, Gmail, Drive. Private to you.