Dwight · Memory, judgment, and agentic follow-through
Try the simulation →Dwight turns scattered work into the next right move.
Dwight reads the places where work actually happens, judges importance versus urgency, remembers your patterns, and stages safe actions across your tools. It is not another empty task list. It is a cockpit for deciding, protecting, delegating, and approving work.
Dwight reads connected work signals and stages actions. You stay the final approval layer.
Builds your matrix
Reads the work signals you connect, turns them into cards, and places each card by importance and urgency.
Explains the judgment
Shows why something is Do Now, Protect, Respond, or Ignore so you can trust, correct, or override it.
Stages next actions
Drafts replies, calendar blocks, source checks, research tasks, and delegation steps before anything executes.
Learns from outcomes
Uses drags, approvals, skips, completed work, and repeated avoidance patterns to score the next day better.
The operating system
One matrix. Multiple surfaces. Same judgment.
Dwight is built around a simple idea: the surface can change, but the memory and judgment should stay consistent. The matrix, chat, cards, Agentic actions, review loops, and mobile capture all point back to the same source-aware model of your work.
Matrix
See what matters now
A live Eisenhower board built from your real calendar, inbox, docs, meetings, time logs, and saved work signals.
Ask Dwight
Reason across your work
Ask what to do, why something matters, what changed, or what Dwight needs before acting.
Card workspace
Move one item forward
Open a card to see evidence, source links, first steps, and safe agentic actions.
Agentic
Let Dwight handle the drag
Stage drafts, blocks, lookups, handoffs, checks, and operator runs in an approval queue.
Review loops
Teach judgment
Approve, skip, drag, finish, or mark stale. Dwight turns those outcomes into memory.
Mobile + companion surfaces
Capture and approve anywhere
Use lightweight surfaces for quick capture, review, and source-aware follow-through.
How to use Dwight
Connect sources
Start with Google. Add the places where work already appears: calendar, mail, docs, Slack, time logs, meeting notes, and other connectors as configured.
Open the matrix
Dwight reads the day and separates what is important from what is merely loud. Do Now, Protect, Respond, and Ignore become the operating map.
Correct the judgment
Drag cards, edit details, or ask why. Your corrections are not just UI changes; they teach the scoring model what you value.
Let Dwight stage the next move
For a card with a clear next action, choose an agentic option: draft, block, verify, research, delegate, open source, or prepare a handoff.
Approve the output
Dwight shows the evidence and proposed action. You approve, edit, skip, or send it to an operator runtime. Nothing sensitive happens silently.
The judgment engine
Dwight uses the Eisenhower Matrix because priority is a trade-off, not a list order.
Every card is placed by two questions: how urgent is it, and how important is it? The quadrant is Dwight's recommendation. You can accept it, drag it, ask why, or turn it into an approved action.
Do Now
Important and time-sensitive. Finish, decide, or unblock today.
Protect
Important before it gets loud. Defend time and make a first move.
Respond
Loud or waiting on you. Batch, draft, delegate, or close.
Ignore
Low value or stale. Drop it, archive it, or let it stay quiet.
Memory, intelligence, judgment
Simple definitions. This is the product model Dwight uses across every surface.
Memory
Dwight remembers the shape of your work: recurring clients, projects, commitments, goals, sources, drags, skips, approvals, and outcomes. Memory is useful only when it improves the next judgment or removes restart friction.
Intelligence
Dwight reads many signals together instead of treating each message or meeting alone. It connects calendar pressure, email asks, documents, meetings, time spent, goals, and past behavior into one view of the day.
Judgment
Judgment is the placement. A card is not just ranked; it is assigned a stance: do it now, protect it, respond efficiently, or ignore it. Dwight explains the trade-off so you can correct it.
The agentic layer
Dwight removes drag without pretending the user disappeared. It prepares work, shows evidence, and asks for approval where it matters.
Draft the reply
Gmail / Slack / WhatsAppDwight reads the thread, writes the likely response, keeps source context attached, and waits for your approval.
Protect the time
CalendarTurns a Protect card into a calendar block with a concrete first move so deep work is easier to enter.
Verify before acting
Stale or risky cardChecks the source, flags uncertainty, and avoids acting on old commitments or ambiguous requests.
Send to Hermes, OpenClaw, Claude Code, Codex, or another runner
Operator handoffDwight remains the memory and judgment layer. Operator tools become the hands that execute scoped work with evidence, approvals, and rollback paths.
Works with operator runtimes
Dwight decides what should happen. Hermes, OpenClaw, and other agents can help do it.
Think of Dwight as the judgment and memory layer. It knows the card, the source, the goal, the evidence, the risk, and the approval state. When a task needs tools, Dwight can package a scoped handoff for an operator runtime and receive a result back for review.
Dwight
Cockpit, memory, judgment, approvals
Turns sources into decisions, keeps the personal context, and decides what is safe or useful to stage.
Hermes / OpenClaw / operator runtimes
Hands and tools
Can receive a scoped task from Dwight, use tools, browse, code, research, or coordinate agents, then return evidence and proposed outputs.
Connectors
Source truth
Google, Slack, Gmail, Calendar, Drive, Granola, Toggl, WhatsApp, GitHub, Linear, and other configured services provide the live signals.
You
Final authority
Approve writes, correct judgments, set goals, and decide when Dwight should act versus wait.
Trust and privacy
The product stance: source-aware, approval-first, honest about uncertainty.
Approval-first writes
Drafts, calendar blocks, handoffs, and outbound actions should be reviewed before execution.
Evidence before confidence
Dwight keeps source links and reasoning close to the card so you can inspect why it recommended something.
Corrections become signal
When you drag, skip, approve, or complete, Dwight treats that as feedback for future scoring.
Frequently asked
What is Dwight in one sentence?
Dwight is an agentic Eisenhower Matrix: it reads your work, judges what matters, stages the next move, and learns from what actually happens.
What does Dwight read?
Dwight can use connected work signals such as Calendar, Gmail, Drive, Slack, Granola meeting notes, Toggl time entries, WhatsApp, GitHub, Linear, and other configured sources. You connect the sources; Dwight builds the matrix from them.
How is Dwight different from a to-do app?
A to-do app waits for you to enter tasks. Dwight finds the tasks in your actual work, explains the priority judgment, and helps stage the next action.
Can Dwight act without me?
Dwight is designed approval-first. It can stage drafts, checks, blocks, handoffs, and operator runs, but sensitive writes should pass through explicit review before they send, post, edit, or schedule.
Calendar, Gmail, Drive, and configured work sources. Private to your Dwight workspace.