Gmail + M365
Provider-neutral email intake and routing
Knowledge + context
Current project state together with the history behind it
Morning Briefing
A realistic daily plan assembled from live operational sources
Human approval
External actions and consequential decisions remain controlled
The starting point
The information needed to plan the day lived in too many places.
In most businesses, that information is spread across email, Microsoft Teams and other chats, calendars, task systems, knowledge bases, and a handful of informal channels. Each tool may do its job well, yet nobody brings the complete picture together. Senior executives can delegate that work to a human assistant. Smaller companies and people in middle management usually cannot—so we built an AI system to do the preparation.
The manual version
Every day started by rebuilding the picture.
Someone had to scan each channel, separate updates from requests, recover the project context, maintain the underlying information, and decide what deserved attention. The work was not a single task; it was the repeated effort of turning scattered signals into a useful plan.
The intervention
Keep the tools. Bring the important signals together.
The assistant collects and filters information behind the existing stack, connects it to the relevant project context, and bundles what remains. People see the emails that need review, the facts that need confirmation, and the decisions that require judgment—instead of processing every input themselves.
The integration principle
Keep the tools. Add the operating layer.
Adding another assistant interface would have recreated the problem in a new place. The system therefore works behind the scenes, while each existing source remains authoritative for the work it already owns.
Gmail and Microsoft 365
Filtering, classification, routing, and project-update detection
Emails queued for review while project information is maintained in the background
Existing calendars
Availability analysis and slot-aware daily planning
Tasks scheduled into free slots and suggested to-dos that fit the time still available
Tasks and to-dos
Business impact, deadline, and required time combined into a dynamic priority
A ranked plan that balances consequential work with what can realistically be completed
Knowledge base and message history
Notion holds current truth; Graphiti links evolving facts, decisions, people, and projects over time
Current project status together with the context that explains how and why it changed
The operating layer maintains context and prepares decisions. It does not create another workspace or message stream for the team to monitor.
The system modules
Four modules turn scattered signals into a workable day.
The modules follow the way work actually arrives: first reduce the noise, then understand the available day, rank the work, and restore the context needed to act with confidence.
01
Information filtering and review
Incoming email is normalised, filtered, and connected to the relevant client or project. Routine information can update the operating picture in the background; anything that needs a response, confirmation, or judgment is placed in a clear review queue.
- Operational categories instead of a generic “important” label
- Requests, actions, and project updates separated from routine traffic
- Drafts and uncertain cases routed to a person before anything is sent
02
Calendar-aware daily planning
The Morning Briefing reads fixed commitments and the free time around them. It proposes work that fits the hours actually available and can place suitable tasks into open calendar slots instead of presenting an impossible wish list.
- Today and tomorrow remain visible
- External deadlines stay separate from general priority
- Focus work and quick tasks are matched to realistic openings in the day
03
Dynamic task priority
A task’s business impact, deadline, and required time remain visible as distinct signals before they are combined into a current overall priority. The result reflects both what matters and what can be done under today’s constraints.
- Stable importance is not confused with temporary urgency
- Required time helps distinguish focus work from useful gap-fillers
- The order can change as deadlines, dependencies, and availability change
04
Project context and organisational memory
The assistant reads the current project record together with the history behind it. It can distinguish an active fact from an older one, recover why a decision changed, and understand how the people involved relate to the work.
- Current state stays explicit and editable
- Decisions, reasons, and changing facts remain retrievable
- Relationships connect people, commitments, and projects instead of returning isolated text
The memory architecture
One place for current truth. Another for how it evolved.
The assistant needs both a working record that people can inspect and a memory that understands changing relationships. We chose Notion and Graphiti for different jobs rather than forcing one system to act as both database and memory.
Operational source of truth
Notion keeps the current state explicit.
Notion combines flexible databases with pages people can read and edit. Its API lets the operating layer query and update structured project information while the team keeps control of the visible source of truth.
- Projects, status, tasks, owners, and deadlines
- Different views over the same structured information
- A human-editable record that automation can maintain through the API
Time-aware relational memory
Graphiti keeps the path behind the state.
Graphiti is an open-source temporal knowledge graph—not a conventional document-only RAG store. It turns messages and events into entities, relationships, and time-stamped facts, so the assistant can follow how projects developed, why decisions changed, and how people are connected to the work.
- New facts can update the current picture without erasing the earlier one
- Relationships connect people, projects, commitments, and decisions
- Source episodes preserve where the contextual knowledge came from
Notion shows the state the team owns. Graphiti supplies the evolving context the assistant needs to interpret it.
Attention by design
A long task list is not a daily plan.
A useful briefing has to answer two different questions: what matters most, and what can realistically be done today. Dynamic priority answers the first. Duration bands make the second visible.
Priority changes with the situation.
A task can be important without being urgent, and urgent without having much long-term impact. The system keeps those signals separate before combining them into a current order of work. The distinction echoes the familiar Eisenhower Matrix, then extends it with deadlines, required time, dependencies, and live availability rather than copying it one-for-one.
01 · Importance
Preserve what matters beyond today
Strategic relevance and expected impact form the stable base. They prevent meaningful work from becoming invisible simply because it is not creating noise yet.
02 · Urgency
Respond to changing operational pressure
Proximity to a due date, dependencies, blocked colleagues, and newly arrived constraints can raise the need for attention as the situation changes.
03 · Deadlines
Keep external commitments explicit
Calendar commitments and external deadlines remain a separate signal in the briefing. They are not buried inside an opaque score that a person has to reverse-engineer.
Five task-duration bands
Priority alone cannot build a workable day. A practical duration estimate lets the briefing distinguish deep focus work from a meaningful task that fits between two meetings.
<15 min
Instant
15–30 min
Quick
<2 hours
Medium
<8 hours
Large
8+ hours
Complex
Capture before prioritisation
A captured thought is not yet a task.
Think of the Inbox like a physical in-tray or a Post-it: it gives a thought, request, or loose end one trusted place to land without interrupting the work in front of you. At the end of the day—or the start of the next—it can be clarified calmly. The context that was only in someone’s head becomes explicit, the thought becomes understandable to another person, and only then is it turned into actionable work. This is a practical adaptation of Getting Things Done, not a one-to-one implementation.
- 01
Capture it quickly
Put the email, thought, request, or open loop into a trusted Inbox without stopping to solve or schedule it on the spot.
- 02
Add the missing context
During a deliberate review, write down what the item means, why it matters, and the relevant background that previously existed only in one person’s head.
- 03
Define the next action
Turn a vague reminder into something another person could understand and execute. If no action is needed, keep it as reference, defer it, or discard it.
- 04
Place and prioritise it
Move the clarified result into the appropriate project, task list, calendar, or waiting queue. Only then can it be ranked against impact, deadlines, required time, and availability.
The Inbox protects focus without hiding unfinished thinking.
Unclarified items remain visible in their own part of the Morning Briefing, so they cannot be forgotten or mistaken for an executable plan. The Inbox comes first in the workflow because it captures context—not because everything inside it deserves first priority.
Read the five-step GTD workflowThe human boundary
Let AI do the preparation. Keep judgment human.
AI is exceptionally useful at collecting evidence, organising context, drafting options, and flagging uncertainty. That is the work this assistant should absorb. It should not quietly send consequential communication or own decisions whose effects a person must answer for. The system reduces the information burden so people can spend their attention on judgment.
Prepare the decision
The system assembles the relevant facts, history, open questions, and possible next steps before asking a person to decide.
Draft before send
The assistant may prepare a response and assemble its context. It does not send consequential external email without explicit approval.
Review uncertainty
Sensitive, ambiguous, or low-confidence cases are surfaced with their sources. The accountable person checks the recommendation and owns the final decision.
Built for client rollout
A reusable system, configured around each business.
This is a production operating system built for repeatable deployment. Its orchestration, memory split, briefing pattern, and control model form the reusable core. Every implementation is then configured around the client’s existing accounts, terminology, responsibilities, policies, and risk boundaries.
- 01
Map the existing flow
Identify the real source systems, communication paths, project ownership, and points where attention is currently lost.
- 02
Configure the context
Adapt classifications, client and project relationships, task fields, duration bands, and briefing priorities to the way the organisation works.
- 03
Agree the boundaries
Define which actions may run automatically, which need review, and how errors, exceptions, permissions, and handover are handled.
Give attention back
Is your team still sorting information—or making the decisions that matter?
If valuable people begin each day by reconstructing context across inboxes, calendars, tasks, and project systems, we can map the flow and build the operating layer that gives that attention back.