AI Assistant
A coach with hands. Every action is logged, every action is reversible. Two modes, 28 tools, memory across conversations.
Why this one is different
Every productivity app has a chat box now. The interesting question isn’t “can it answer?” — it’s “can you trust it to act?”
CoachKeeper answers that question the same way every time:
- Every action streams into a live log as it happens — no hidden side effects.
- Every action is reversible — an Undo button on the log entry, or just ask the assistant to undo it.
- It honors the same rules you do — your availability windows, the auth boundary, the data shapes.
That’s what makes the rest of this page interesting: the agent has hands because we made having hands safe.
Two modes
The AI has two personalities, switchable anytime from the chat header:
Assistant mode
Direct and task-focused. You say what you need, it does it. No extra commentary, no unsolicited advice. Ideal when you know what you want and just need it executed.
Coach mode
A thoughtful productivity partner. Coach mode opens with a brief check-in on your current state, how many tasks are overdue, what’s been sitting in progress too long, whether your priorities have scheduled time. It surfaces patterns, suggests what to focus on, and gently nudges you forward.
The tone is always an offer, never a guilt trip: “Shall we tackle…” not “You still haven’t…”. Think of it as a thoughtful colleague who happens to have perfect memory.
Three high-impact things to ask first
Skip the feature dump for a minute. These three prompts give you the clearest sense of what the coach actually does:
“What should I focus on today?” — it reads your backlog, your calendar, your availability, and helps you see where to start. Real briefing, not a horoscope.
“Turn the action items in [@this note] into todos and schedule them this week.” — meeting notes become real tasks on real days, in seconds.
“What did that research paper say about retention?” — semantic search over your uploaded sources, with the exact passage cited back to you. Citations appear as pills under the answer; click one to open the source in the in-app viewer, right at the cited page.
If those three work for you, the rest of the toolset is just expansion. The full surface, grouped by domain:
| Domain | Capabilities |
|---|---|
| Notes | Create, update, delete, search, summarize, rewrite |
| Todos | Create, update, change status, modify checklists, search |
| Events | Create, update, delete, check availability, handle recurring events |
| Sources | List uploads, read content, search by meaning |
| Search | Find items across all entity types by title or content |
| Scheduling | Read your availability windows, suggest time blocks |
| Web search | Search the internet (when enabled in settings) |
| Memory | Learn and recall your preferences across conversations |
| Undo | Reverse any recent action, even multiple at once |
Most of this toolset is also available to external AI clients — point Claude Code, Claude Desktop, or Cursor at your workspace via the MCP Integration guide.
How to talk to it
Just type naturally. The assistant understands context, it knows what tab you have open, what’s in your backlog, and what’s on your calendar.
Everyday prompts:
“What should I focus on today?”
“Create a todo to review the quarterly report and schedule an hour for it tomorrow morning.”
“Find my notes about the marketing strategy and summarize the key points.”
“What’s been in progress for more than a week?”
“Move my 2pm meeting to Thursday.”
Batch operations:
“Create todos for each action item in my meeting notes.”
“Schedule my top 3 backlog items this week.”
Knowledge queries:
“What did that research paper say about retention rates?”
“Summarize my notes from last Tuesday.”
When an answer draws on your notes or uploaded sources, citation pills appear beneath it showing exactly what was used. Pills that point at a source are clickable — one click opens the document in the in-app viewer at the cited page, so you can verify the passage yourself.
Quick prompts
Pre-built prompts to get started fast:
- “Plan my day”, reviews your backlog and calendar, suggests a plan for the rest of the day
- “Schedule my week”, creates time blocks across your available hours
- “What’s overdue?”, lists tasks that have been sitting too long
- “Weekly review”, summarizes what you completed, what’s in progress, and what’s coming up
@Mentions
Type @ in the chat input to reference a specific note, todo, event, or source. An autocomplete popup helps you find the right item. The AI pins that entity to the conversation with its full content, so it has deep context without you needing to describe what you’re talking about.
Mentioning an uploaded file by name — @report.pdf — resolves to the actual document, and the assistant can search inside its extracted text to answer questions about it, not just skim the title.
The action log
Every action the AI takes appears in a live action log within the chat. You see entries like:
- CREATE Todo: “Review quarterly report”
- CREATE Event: “Deep Work: Q Report”, Thursday 9:00–10:00 AM
- UPDATE Todo status: in_progress → done
Each entry has an “Open” button to jump directly to the affected item, and an Undo button to reverse it.
This transparency is a core design principle. You always see what changed.
Undo
Every AI action can be reversed. Click the Undo button on the entry in the action log, or ask the AI: “Undo that” or “Undo the last 3 actions.” Up to 20 recent actions are tracked for reversal.
For bulk operations (six or more changes), the AI automatically presents its plan and waits for your approval before executing. No surprises.
Context awareness
The assistant automatically knows:
- Your open tab, if you have a note open and say “summarize this,” it knows which note
- Your task state, backlog count, in-progress items, estimated hours of work
- Your calendar, upcoming events, free time slots
- Your availability windows, it never schedules outside your working hours
- Your uploaded documents, searchable by meaning via semantic search
- Your preferences, learned over time through its memory system
Memory
The AI learns your patterns and preferences over time:
- How you like things scheduled (morning focus blocks, afternoon meetings)
- Your naming conventions and recurring projects
- Communication preferences (concise vs. detailed responses)
Memories persist across conversations. You can view and delete them anytime in Settings > AI Memory.
Tips
- Be specific. “Schedule 2 hours tomorrow morning” works better than “schedule something.”
- Try Coach mode when you feel scattered, the check-in and gentle nudges help you regain focus.
- Use @mentions for long notes or documents, it saves you from describing them.
- Ask it to plan, not just create. “Plan my week based on my backlog” gives better results than creating events one by one.
- Trust the undo. If the AI creates something wrong, one click fixes it.